oli framework case study

• Recently Active
• Top Discussions
• Best Content

By Industry

• Investment Banking
• Private Equity
• Hedge Funds
• Real Estate
• Venture Capital
• Asset Management
• Equity Research
• Investing, Markets Forum
• Business School
• Fashion Advice
• Technical Skills
• Strategy Resources

Eclectic Paradigm

A three-tiered review framework to determine if a Foreign Direct Investment (FDI) is advantageous.

• What Is The Eclectic Paradigm?
• Understanding An Eclectic Paradigm
• Ownership Advantage 

Location Advantages

Internalization advantage.

• Example Of An Eclectic Paradigm (OLI Framework)
• Dunning's Four Motives For FDI
• Extension Of The OLI Model

  What Is The Eclectic Paradigm?

An eclectic paradigm, also known as the ownership, location, internalization (OLI) model or OLI framework, is a three-tiered review framework that companies can use to determine if a Foreign Direct Investment (FDI) is advantageous.

The eclectic paradigm predicts that institutions will avoid open market transactions if internal or in-house completion of identical work is less expensive. It is based on internalization theory and was first formulated in 1979 by researcher John H. Dunning.

By relocating certain value chain activities overseas, firms can expand their operations outside national borders. There are several reasons why businesses do so. 

First, they may seek more desirable natural or strategic resources, including physical, financial, technological, and human capital. In other circumstances, the company's sole objective may be to gain new customers.

It may export more goods and services than it sells domestically. They may choose to operate in the nation that purchases their goods the most, avoiding shipping fees. 

To achieve economies of scale and reduced costs, firms seek to improve efficiency. 

Regardless of the benefits, these companies need a complete international expansion strategy.

Key Takeaways

• The Eclectic Paradigm, also known as the OLI Model, is a framework for understanding the factors that drive multinational enterprises (MNEs) to engage in foreign direct investment (FDI).
• The paradigm is based on three key components: Ownership advantages (O), Location advantages (L), and Internalization advantages (I). These components help explain why companies choose to invest abroad rather than exporting or licensing.
• The Eclectic Paradigm is used by businesses to decide whether, where, and how to invest abroad. It provides a comprehensive framework for evaluating the strategic rationale behind FDI decisions.
• Critics argue that it does not account for the complexities and rapid changes in the global business environment. Nonetheless, it remains a foundational theory in international business studies.

Understanding an Eclectic Paradigm

A commercial and economic evaluation is important when determining the feasibility of foreign direct investment . There are numerous entry-mode tactics, each with its advantages and disadvantages.

Common investment strategies include exporting , licensing, franchising, building a strategic alliance, establishing a joint venture , acquiring, and beginning from scratch with a greenfield investment. 

The latter three examples are  foreign direct investments  because they involve substantial stock investments. 

Utilizing the OLI framework effectively excludes at least a portion of these options (this is also known as the eclectic paradigm). 

According to this paradigm, a company must possess three benefits to engage in FDI successfully. If these criteria aren't fully met, the company may decide to utilize an alternative entry-mode strategy. 

The basic hypothesis of this paradigm is that the level and structure of a company's international value-adding activities will be contingent on meeting three parameters:

• Exclusive and enduring ownership of certain advantages relative to other companies (Ownership Advantages – "O")
• The extent to which the company believes it is advantageous to add to its O advantages as opposed to selling them to other overseas companies (Internalization Advantages – "I")
• The amount firms are interested in generating, gaining access to, or leveraging their "O" advantages in a foreign place (Location Advantages - "L"). 

Ownership Advantage 

To overcome the disadvantage of foreignness, a firm must first possess an ownership advantage. 

The liability of foreignness refers to the inherent disadvantage foreign businesses suffer in host countries due to their non-native status. These disadvantages range from simple linguistic difficulties to unfamiliarity with local client requirements.

A company's ownership advantages include its proprietary information and other ownership rights. This advantage results from the company's brand, copyright, trademark, or patent rights, as well as its internal use and expertise. 

Consequently, ownership benefits are typically perceived as intangible. These benefits should be attractive, rare, difficult to imitate, and engrained in the organization. 

In other words, the resource must be valuable enough to provide a competitive edge over an international rival.

The OLI paradigm does not restrict "O" benefits to firm-specific benefits. Instead, they pertain to competencies that a business can cultivate internally or acquire externally through business networks.

Dunning divided ownership benefits into asset-based (Oa) and transaction-based (Ot) benefits.

He indicated that the OLI Paradigm might be utilized in studies examining phenomena such as FDI effects, foreign disinvestment, and FDI-related dynamics, among others.

Therefore, businesses should evaluate if they possess a competitive advantage that they can export to compensate for their foreignness liability. 

Examples include a powerful brand name with a stellar reputation, exclusive technology or skills, and significant economies of scale .

The second essential advantage is the location advantage. Considering the risk of foreignness, host countries must offer compelling benefits to make internationalization profitable for FDI. 

These advantages may be merely geographical or the result of inexpensive raw resources, low salaries, a skilled workforce, special taxes, the absence of tariffs, etc. 

Dunning expanded on the L component upwards (regional integration) and downwards (clustering), demonstrating his interest in the subnational level.

Companies must determine if there is a comparative benefit to conducting certain services in a given nation. Frequently, these factors depend on the cost and availability of resources. 

Moreover, the characteristics of the selected areas vary. Location advantage typically refers to natural or artificial resources. 

These resources are typically immobile and require collaboration with a foreign investor in the target site to be utilized to their maximum potential.

"L" benefits are not limited to country-specific benefits. In addition, they include benefits associated with supra-national regions and geographic advantages inside a host country, particularly at the sub-national level, as found in regional clusters.

The Porter's Diamond model is an excellent instrument for identifying these location advantages. Companies should consider whether the target market offers any location advantages. 

To decide between licensing and FDI management, the eventual advantage of internalization should be considered. For example, is it preferable to do the value chain activity internally rather than contracting it out? 

Certain activities may be outsourced to foreign businesses because they are better at doing it for less money, have greater local market expertise, or because management wishes to focus on other activities in the value chain, such as marketing or design.

Lastly, internalization advantages indicate when an organization should create a specific product internally instead of contracting with a third party. 

Occasionally, it may be more economical for a company to run from a separate market location while retaining production in-house.

If a business chooses to outsource production, it may be necessary to negotiate relationships with local suppliers.

However, outsourcing is only economically viable if the contracting company can meet the organization's policies, standards, and quality criteria at a much lower price. 

As an alternative, the foreign partner may provide knowledge of the local market or even better workers who may produce a superior product that the company lacks.

The best internationalization options for a corporation to employ include forming joint ventures with local partners, buying already-existing local businesses, or beginning from scratch with a greenfield investment.

Example of an Eclectic Paradigm (OLI Framework)

Businesses can use the OLI framework to assess, comprehend, and decide how much foreign direct investment (FDI) to make. 

For instance, a business considering a foreign investment may utilize the OLI framework to identify the most important ownership, location, and internalization aspects that will influence its choice.

Maybe the business is looking for a place with a low labor cost. In this situation, the business would prefer to find a nation with a cheap cost of ownership. 

Alternatively, the business may look for a nation with a skilled labor force. In this situation, the business should look for a nation with a high internalization cost.

The company may choose where to invest more wisely and strategically if it knows the criteria that matter most. Using the OLI methodology, researchers can also analyze FDI and forecast foreign investment trends.

The patterns of foreign direct investment can be better understood and forecasted by studying the ownership, location, and internationalization elements that affect a company's decision-making. 

Then, based on these forecasts, organizations and governments can create policy recommendations.

Case study of Shanghai Vision Technology Co. Ltd.

According to Research Methodology, independent research and analysis firm  Shanghai Vision Technology Company  utilized this paradigm while deciding to export its 3D printers and other unique technologies. 

While they carefully considered the disadvantages of higher tariffs and transportation expenses, their internationalization approach ultimately enabled them to thrive in new markets.

Assessing the Eclectic Paradigm

The OLI paradigm claims that the specific configuration of the OLI parameters facing a given firm and the firm's response to that configuration is highly contextual. 

In particular, it will reflect the economic and political characteristics of the investment firms and the country or region in which they seek to invest.

The industry and nature of the value-added activity in which the firms are engaged; the characteristics of the individual investing firms, including their objectives and strategies for pursuing these objectives; and the reason for the FDI.

Combining our knowledge of the OLI parameters, the economic features, geographical features, etc. of the home and host countries, and the investing or potentially investing enterprises, we may generate a wide range of reasonably detailed and operationally testable ideas. 

Thus, it may be hypothesized that some industries, such as the oil and pharmaceutical industries, are more likely to generate FDI than others, such as the iron and steel or aircraft industries.

This is because the characteristics of the former generate more unique O advantages, whether for their locational needs favoring production outside their home countries or because the net benefits of internalizing cross-border intermediate products are greater.

Dunning's Four Motives for FDI

Incorporating insights from strategic management and basing his findings on the distinction between Oa and Ot, Dunning identified four reasons for FDI:

1. Market-Seeking: To capitalize on the opportunities given by larger markets.

(i) Gain access to new markets/opportunities

→ Local production in other nations enhances customer service and lowers the price of shipping.

(ii) Follow key customers

→ Businesses may follow important clients abroad to prevent other suppliers from providing their services. Additionally, the company is better positioned to meet customer needs by starting local businesses.

(iii) Compete with critical rivals in their markets

→ To damage a competitor by making it use resources to protect its market, it is advantageous to compete against present or potential rivals in their home market.

2. Efficiency-Seeking: To take advantage of variances in the cost and availability of traditional factor endowments, economies of scale and scope in various nations are utilized.

(i) Economies of scale and scope 

→ The company can boost its economies of scale by going global. This is all made possible by falling fixed costs , workforce specialization, bulk discounts, and access to other financial economies.

→ Economies of scope are the cost reductions that result from utilizing a relatively fixed base of managerial personnel, infrastructure, and other business resources across a broad suite of products and services.

(ii) Locate production near customers

(iii) Take advantage of government incentives

(iv) Avoid trade barriers

3. Strategic Asset-Seeking Investment: Instead of leveraging existing assets, acquire and supplement a new technical basis.

(i) Knowledge

→Through FDI, a company can create a local presence and better understand its target markets. FDI typically gives the foreign company better access to market information, clients, distribution networks, and management of local businesses.

(ii) Technological & managerial know-how

→ The company might gain from having a presence in a significant industrial hub. By doing this, the business will be surrounded by organizations that share its goals and will be encouraged to innovate and advance.

4. Resource-Seeking Investment: To gain access to specific types of resources that are unavailable or more costly in the native country.

(i) Raw materials needed in extractive and agricultural industries

→Businesses in the agricultural and extractive sectors are forced to go to the locations of the raw materials.

(ii) Unskilled/semi-skilled labor 

→ Companies may decide to use FDI to shift production to low-wage nations to reduce costs and boost profit margins.

These factors pertain to selecting the optimal location for every FDI project.

Extension of the OLI Model

The OLI paradigm's analytical framework links various theories of international business) derived from concepts from various disciplines. 

It can be applied at various levels of analysis to understand the types of foreign production, the contexts in which they exist, and why multinational enterprises (MNEs) choose them.

Business groups don't typically have the "firm-specific advantages" (FSAs) that the OLI framework mentions, but their ownership advantages come from the locational advantages of their home country. 

FSAs have significant brand recognition, IP rights, and other intangible asset benefits. Because ownership and locational advantages interact and reinforce one another rather than acting alone, the FSAs of business groups are characterized by the locational advantages of their home countries.

The FSAs of corporate groups will give "reduced marginal gains" in their internationalization process as pro-market reforms are enacted in emerging economies, strengthening their integration with the global economy .

These FSAs, however different from the classic multinational FSAs of industrialized countries and will offer the same benefits as those covered by the Ownership-Location-Internalization (OLI) paradigm.

As a result, certain additions have been made to the OLI Paradigm, resulting in the OLMA model (meaning Ownership, Location, Mode of entry, and Adjustment). 

Others have interpreted the paradigm in e-business and given the OLI configuration a network-based advantages component.

Some have added a political perspective to the OLI paradigm to explain FDI in fragile and emerging frontier markets.

Everything You Need To Master Financial Statement Modeling

To Help you Thrive in the Most Prestigious Jobs on Wall Street.

Free Resources

To continue learning and advancing your career, check out these additional helpful  WSO  resources:

• Mission Statement
• Price Skimming
• Diversification
• Pareto Analysis
• Strategic Analysis

Get instant access to lessons taught by experienced private equity pros and bulge bracket investment bankers including financial statement modeling, DCF, M&A, LBO, Comps and Excel Modeling.

or Want to Sign up with your social account?

Eclectic Paradigm

An economic and business method for analyzing the attractiveness of making a foreign direct investment

What is the Eclectic Paradigm?

Based on the internalization theory of British economist J.H Dunning, the eclectic paradigm is an economic and business method for analyzing the attractiveness of making a foreign direct investment (FDI) . The eclectic paradigm model follows the OLI framework. The framework follows three tiers – ownership, location, and internalization.

The eclectic paradigm assumes that companies are not likely to follow through with a foreign direct investment if they can get the service or product provided internally and at lower costs.

• The eclectic paradigm is an economic and business method for analyzing the attractiveness of making a foreign direct investment.
• The eclectic paradigm model follows the OLI framework. The framework follows three tiers – ownership, location, and internalization.
• Ownership can be defined as the proprietorship of a unique and valuable resource that cannot easily be imitated, which creates a competitive advantage against potential foreign competitors.

Ownership Advantage

The ownership advantage can also be seen as the competitive advantage that comes with the FDI. Ownership, in this instance, can be defined as the proprietorship of a unique and valuable resource that cannot easily be imitated, thereby creating a competitive advantage against potential foreign competitors.

The intrinsic disadvantages or challenges associated with FDIs, in terms of ownership, circle around the liabilities that come with foreignness since the potential investor is a non-native in the country that the FDI will be made. The challenges can include (but are not limited to) possible language barriers or lack of knowledge of the demand trends that are common among the local consumer markets.

Companies and their management teams normally need to consider the possibilities of transference of the competitive advantage to other foreign markets in order to counterbalance the liabilities mentioned above. Ideally, an attractive investment should include notable economies of scale , a sound reputation, and a well-known brand name, advanced technology, etc.

Location Advantage

The potential business host countries being considered for FDIs must present numerous competitive advantages; location is one of them. The location advantage focuses more on the geographic advantages of the host country or countries. An example of a geographic advantage can be access to the ocean (for sea freight or other purposes) versus a land-locked country.

Other location advantages can include low-cost labor and raw materials, lower taxes and other tariffs, a well-trained labor force, etc. Normally, the Porter’s diamond model can be used to evaluate location advantages.

Companies and their management teams normally need to consider whether any location advantages, as mentioned above, exist in the market they wish to enter. Should the advantages exist, the companies can consider taking on the investment through an FDI or other pathways (e.g., franchising or licensing) provided that there is a demand in the foreign markets.

Internalization Advantage

In order for companies to choose which investment pathway or method is best suited for their needs, their management team must analyze the internalization advantage. They normally need to consider whether it would be more sensible to get the value chain activity performed locally with their own team or outsource it to a foreign country.

The advantages of outsourcing from different countries can include (but are not limited to) lower costs and better skills to perform the value chain activities and/or better knowledge of the local markets.

In such a case, management can choose between two options on how to proceed. It can either outsource its production to an original equipment manufacturer (OEM) or license its product design to an independent foreign company

If the company does so, however, it should keep control over its activities and engage in FDI. It can be done by starting from scratch through a greenfield investment, entering into joint ventures with local partners, or purchasing existing local companies.

More Resources

CFI is the official provider of the global Commercial Banking & Credit Analyst (CBCA)™ certification program, designed to help anyone become a world-class financial analyst. To keep advancing your career, the additional CFI resources below will be useful:

• Greenfield Investment
• Outsourcing
• Value Chain
• See all management & strategy resources
• Share this article

Create a free account to unlock this Template

Access and download collection of free Templates to help power your productivity and performance.

Already have an account? Log in

Supercharge your skills with Premium Templates

Take your learning and productivity to the next level with our Premium Templates.

Upgrading to a paid membership gives you access to our extensive collection of plug-and-play Templates designed to power your performance—as well as CFI's full course catalog and accredited Certification Programs.

Already have a Self-Study or Full-Immersion membership? Log in

Access Exclusive Templates

Gain unlimited access to more than 250 productivity Templates, CFI's full course catalog and accredited Certification Programs, hundreds of resources, expert reviews and support, the chance to work with real-world finance and research tools, and more.

Already have a Full-Immersion membership? Log in

To read this content please select one of the options below:

Please note you do not have access to teaching notes, the oli paradigm as a comprehensive model of fdi determinants: a sub-national approach.

International Journal of Emerging Markets

ISSN : 1746-8809

Article publication date: 14 September 2020

Issue publication date: 21 January 2022

This study aims to advance a sub-national perspective within the OLI Paradigm by analyzing how and to what extent the Eclectic Paradigm can serve as a general model to capture region-specific aspects of the location determinants of FDI, encompassing institutional effects that extend beyond the quality of institutions.

Design/methodology/approach

The authors conducted a systematic literature review of 41 selected papers published between 1990 and 2019. Using inductive content analysis, they investigated the theoretical choices used to support analyses of the effects of institutional factors on MNEs' location decisions at the sub-national level.

It was found that, when changing from the national to the sub-national level of analysis, there is no need to change the main assumptions used in the literature, although a different perspective must be adopted. The Eclectic Paradigm permeates most of the studies revised and can serve as a general model to capture the sub-national perspective. It offers a foundation for new perspectives on the dynamics of institutional and political factors and their effects on location strategies and determinants at the sub-national level. Adopting the OLI Paradigm with a sub-national approach could widen the IB literature's prevailing focus on traditional economic factors and institutional quality.

Research limitations/implications

The authors contribute to extant International Business literature Their paper enhances the literature on FDI location determinants by providing a more specific approach to development of a sub-national perspective within the OLI Paradigm, extending the institutional effects to capture more region-specific factors influencing the location of FDI. Study limitations are related to our analytical focus on the location dimension, excluding motives for FDI or firm-level location strategies. Rather than limiting analysis to quantitative studies, future research that includes qualitative studies and also covers the other dimensions of the OLI Paradigm could open additional new research avenues for advancing the sub-national perspective within the field of IB.

Practical implications

The authors’ main findings suggest that MNEs' location strategies should include a sub-national perspective, which means that firms need to assess different levels of the location and understand their interaction with nationwide constraints and limitations, as it may affect firms' ability to effectively conduct their value-adding activities. They also contribute elements that can support sub-national governments' actions and policies aiming to enhance locational advantages to attract and retain FDI.

Originality/value

This review specifically analyzes the location determinants of FDI at the sub-national level, in studies published in a broad set of journals, from a variety of fields, prioritizing articles that investigate sub-national institutional determinants. The authors derive implications for the International Business literature and propose that the sub-national dimension should be incorporated into the Eclectic paradigm in order to better understand the influence of institutional sub-national determinants.

• Foreign direct investment (FDI)
• Institutions
• Sub-national
• Eclectic paradigm

Acknowledgements

Ministério da Ciência, Tecnologia e Inovação > Conselho Nacional de Desenvolvimento Científico e TecnológicoThis research has been supported by the Brazilian National Council for Scientific and Technological Development (Edital Universal 28/2018).

Batschauer da Cruz, C.B. , Eliete Floriani, D. and Amal, M. (2022), "The OLI Paradigm as a comprehensive model of FDI determinants: a sub-national approach", International Journal of Emerging Markets , Vol. 17 No. 1, pp. 145-176. https://doi.org/10.1108/IJOEM-07-2019-0517

Emerald Publishing Limited

Copyright © 2020, Emerald Publishing Limited

Related articles

All feedback is valuable.

Please share your general feedback

Report an issue or find answers to frequently asked questions

Contact Customer Support

The browser you are using is not supported by this website. All versions of Internet Explorer are no longer supported, either by us or Microsoft (read more here: https://www.microsoft.com/en-us/microsoft-365/windows/end-of-ie-support ).

Please use a modern browser to fully experience our website, such as the newest versions of Edge, Chrome, Firefox or Safari etc.

Internationalization of Chinese MNEs and Dunning’s Eclectic (OLI) Paradigm:A Case study of Huawei Technologies Corporation’s Internationalization Strategy

• Beiguang Zhu

Summary, in English

Department/s

• Department of Economics

Publishing year

• Available as PDF - 472 kB
• Download statistics

Document type

Student publication for Master's degree (one year)

• Business and Economics
• internationalization
• Chinese MNE
• Dunning Eclectic Paradigm
• Economics, econometrics, economic theory, economic systems, economic policy
• Nationalekonomi, ekonometri, ekonomisk teori, ekonomiska system, ekonomisk politik
• Sonja Opper

Academia.edu no longer supports Internet Explorer.

To browse Academia.edu and the wider internet faster and more securely, please take a few seconds to  upgrade your browser .

Enter the email address you signed up with and we'll email you a reset link.

• We're Hiring!
• Help Center

ANALYSIS OF MOTIVES AND PROSPECTS WITHIN THE OLI FRAMEWORK: A CASE STUDY OF GERMAN FDI IN CHINA, By CHINWENDU BLESSING OKOLI. Trade and Multinational Enterprises

Related Papers

Katiuscia Vaccarini

The purpose of this systematic literature review is to investigate whether and to what extent psychic distance (PD) is considered by scholars as a specific determinant of inbound and outbound foreign direct investments (FDI) in China. Main findings show that PD as a determinant of FDI seems to be under-investigated. The difficulty in quantifying such a subjective determinant in international business (IB) research area might be one of the main reasons. Nevertheless, the literature does raise awareness on "distance" and its related multi-dimensions since the concept first appeared in the 1950s. The most applicable indexes elaborated over the years are based on Hofstede’s work. These indexes are not only applied in IB but also in psychology and management, among the others and this pinpoints the relevance of PD and its interdisciplinary nature. More specifically, this paper aims to systematize the limited and fragmented literature about non-tangible perceptions of distance between home and host countries. This work contributes in raising awareness on the topics of PD and cultural distance. However, as part of a wider European Union project the ultimate results of the paper have not been leaked yet. This study provides useful insights and practical implications at three levels: international scholars (a more comprehensive review and systematization of the concept of PD), managers (awareness raising on PD, foreign environment opportunities and challenges, setting up and strengthening long-term cooperation between Europe and China). Finally, policy makers (trade and investment flows between Europe and China, key barriers which affect companies and require action politically and institutionally to ensure that opportunities are fully exploited by companies).

Marian Gorynia

Alex Kostin

This article describes the origins, and traces the subsequent evolution of the eclectic paradigm from the mid-1950s to the present day. It does so in the light of the changing characteristics of MNE activity and of the global economic scenario. The article concludes by asserting that the eclectic paradigm still remains a powerful and robust framework for examining contextual specific theories of foreign direct investment and international production.

Foreign direct investment in the service sector has been gaining importance in the past decade as more countries transform themselves into post-industrialised economies. The transition from a centrally planned economy to market-based economy has generated a surge of foreign direct investment from industrialised countries to Central and Eastern European Countries. This paper examines the impact of ownership and location factors on the extent of internalisation for service multinationals seeking to enter into the Czech Republic, Hungary, and Poland, as these countries launch their economies towards increased privatisation, deregulation, and liberalisation. Using foreign direct investment data of 76 firms during 1990–2000, we find significant support for our main hypotheses.

Piotr Trapczynski

This study attempts to investigate the establishment mode choice of Polish firms when entering foreign markets via foreign direct investment (FDI) in the light of J. Dunning’s eclectic OLI model. A dichotomous mode choice between green-field investment and acquisitions forms the basic framework for analysis. The theoretical part contains the application of Dunning’s model to the investigated relationships with the said establishment mode choice as well as an overview of empirical research conducted so far on outward foreign direct investment (OFDI) from the countries of Central and Eastern Europe (CEE). Seven research hypotheses are formulated in this context and tested on a sample of firms registered in Poland with ongoing FDI. Data collection was made using an online survey, then four models were used to verify the statistical significance of the hypothesized relationships. The results showed no statistical significance for hypotheses linking mode choice with ownership and interna...

DINKAR N NAYAK

Abhinav Gokari

Procedia - Social and Behavioral Sciences

anusree paul

Phong Kinh tế Đối ngoại Sở Kế hoạch và Đầu tư, TT Huế

Huy Trần Đăng

The paper represents the author's attempt to stress the importance of foreign direct investment (FDI) as the key factor of economic development, and basic mechanism of capital flows, more desirable than bank debts or portfolio equity investment. The fundamental premise of Dunning's eclectic paradigm or the OLI model is that returns on foreign investment as a basic motive for FDI can be explained by three groups of factors: the ownership advantage of the firm (O), location factors (L), and by internalisation of trasaction costs (I). Since we can assume that foreign investors already posses certain competitive (ownership) advantage, and they are able to internalize transaciton costs, the key remaining factor in decision-making process are the location advantages of the host country. There is a bulk of location determinants, and among them the author especially points out the institutional factors, which derive from the FDI policy regime of a country.

Loading Preview

Sorry, preview is currently unavailable. You can download the paper by clicking the button above.

RELATED PAPERS

Arthur Poon

Andrew uche

Renée S Hunter

PYRROS PAPADIMITRIOU

International Business Review

Peter J. Buckley

•   We're Hiring!
•   Help Center
• Find new research papers in:
• Health Sciences
• Earth Sciences
• Cognitive Science
• Mathematics
• Computer Science
• Academia ©2024

International business and the eclectic paradigm: developing the OLI framework

• Book Review
• Published: 29 September 2004
• Volume 35 , pages 456–458, ( 2004 )

Cite this article

• Ram Mudambi 1  

4362 Accesses

6 Citations

Explore all metrics

This is a preview of subscription content, log in via an institution to check access.

Access this article

Subscribe and save.

• Get 10 units per month
• Download Article/Chapter or eBook
• 1 Unit = 1 Article or 1 Chapter
• Cancel anytime

Price includes VAT (Russian Federation)

Instant access to the full article PDF.

Rent this article via DeepDyve

Institutional subscriptions

Brouthers, L.E., Brouthers, K.D. and Werner, S. (1999) ‘Is Dunning's eclectic framework descriptive or normative?’, Journal of International Business Studies 30: 831–844.

Article   Google Scholar  

Casson, M.C. (1994) ‘Institutional diversity in overseas enterprise: Explaining the free-standing company’, Business History 36(4): 95–108.

Dunning, J.H. (1988) Explaining International Production, Unwin Hyman: London.

Google Scholar  

Hoschka, T. and Livingston, J. (2002) ‘Winning Asian strategies’, The McKinsey Quarterly 1: 6–9.

Itaki, M. (1991) ‘Critical assessment of the eclectic theory of the MNE’, Journal of International Business Studies 22(3): 445–460.

Mudambi, R. (1999) ‘MNE internal capital markets and subsidiary strategic independence’, International Business Review 8(2): 197–211.

Robins, J., Tallman, S. and Fladmoe-Lindquist, K. (2002) ‘Autonomy and dependence in of international cooperative ventures: an exploration of the strategic performance of US ventures in Mexico’, Strategic Management Journal 23: 881–901.

Download references

Author information

Authors and affiliations.

Department of General & Strategic Management, Fox School of Business, Temple University, Philadelphia, PA, USA

Ram Mudambi ( Reviewed by: )

You can also search for this author in PubMed   Google Scholar

Rights and permissions

Reprints and permissions

About this article

Mudambi, R. International business and the eclectic paradigm: developing the OLI framework. J Int Bus Stud 35 , 456–458 (2004). https://doi.org/10.1057/palgrave.jibs.8400092

Download citation

Published : 29 September 2004

Issue Date : 01 September 2004

DOI : https://doi.org/10.1057/palgrave.jibs.8400092

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Find a journal
• Publish with us
• Track your research

•  Sign into My Research
•  Create My Research Account
• Company Website
• Our Products
• About Dissertations
• Español (España)
• Support Center

Select language

• Bahasa Indonesia
• Português (Brasil)
• Português (Portugal)

Welcome to My Research!

You may have access to the free features available through My Research. You can save searches, save documents, create alerts and more. Please log in through your library or institution to check if you have access.

Translate this article into 20 different languages!

If you log in through your library or institution you might have access to this article in multiple languages.

Get access to 20+ different citations styles

Styles include MLA, APA, Chicago and many more. This feature may be available for free if you log in through your library or institution.

Looking for a PDF of this document?

You may have access to it for free by logging in through your library or institution.

Want to save this document?

You may have access to different export options including Google Drive and Microsoft OneDrive and citation management tools like RefWorks and EasyBib. Try logging in through your library or institution to get access to these tools.

• More like this
• Preview Available

Foreign direct investment: the OLI framework

No items selected.

Please select one or more items.

Select results items first to use the cite, email, save, and export options

You might have access to the full article...

Try and log in through your institution to see if they have access to the full text.

Content area

(ProQuest: ... denotes formulae omitted.)

The OLI, or "eclectic," approach to the study of foreign direct investment (FDI), developed by John Dunning, has proved to be an extremely fruitful way of thinking about multinational enterprises (MNEs) and has inspired a great deal of applied work in economics and international business. In itself it does not constitute a formal theory that can be confronted with data in a scientific way, but it nevertheless provides a helpful framework for categorizing much (diough not all) recent analytical and empirical research on FDI. This survey first summarizes the OLI paradigm and then uses it as a lens through which to review some of the highlights of this research, while also noting some important issues that it neglects.

OLI stands for ownership, location, and internalization, three potential sources of advantage that may underlie a firm's decision to become a multinational. Ownership advantages address the question of why some firms but not others go abroad, and suggest that a successful MNE has some firm-specific advantages that allow it to overcome the costs of operating in a foreign country. Location advantages focus on the question of where an MNE chooses to locate. Finally, internalization advantages influence how a firm chooses to operate in a foreign country, trading off the savings in transactions, hold-up, and monitoring costs of a whoUy owned subsidiary against the advantages of other entry modes such as exports, licensing, or joint venture. A key feature of this approach is that it focuses on the incentives facing individual firms. This is now standard in mainstream international trade theory but was not at all so in the 1970s, when FDI was typically seen through a Heckscher-Ohlin lens as an international movement of physical capital in search of higher returns (see, for example, Mundell 1957).

Ownership Ownership advantages are key to explaining the existence of MNEs. A central idea is that firms are collections of assets, and that candidate MNEs possess higher-than- average levels of assets having the character of internal public goods. These assets can be applied to production at different locations without reducing their effectiveness. Examples include product development, managerial structures, patents, and marketing skills, all of which are encompassed by Helpman's (1984) catchall term headquarter services. Although this...

You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer

Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer

Suggested sources

• About ProQuest
• Terms of Use
• Privacy Policy
• Cookie Policy

• Search Search Please fill out this field.
• Corporate Finance

Eclectic Paradigm: Definition, Example, Advantages

James Chen, CMT is an expert trader, investment adviser, and global market strategist.

What Is an Eclectic Paradigm?

An eclectic paradigm, also known as the ownership, location, internalization (OLI) model or OLI framework, is a three-tiered evaluation framework that companies can follow when attempting to determine if it is beneficial to pursue foreign direct investment (FDI). This paradigm assumes that institutions will avoid transactions in the open market if the cost of completing the same actions internally, or in-house, carries a lower price. It is based on internalization theory and was first expounded upon in 1979 by the scholar John H. Dunning.

Key Takeaways

• An eclectic paradigm is also known as the ownership, location, internalization (OLI) model or OLI framework.
• The eclectic paradigm takes a holistic approach to examining entire relationships and interactions of the various components of a business.
• The goal is to determine if a particular approach provides greater overall value than other available national or international choices for the production of goods or services.

Understanding Eclectic Paradigms

The eclectic paradigm takes a holistic approach to examining entire relationships and interactions of the various components of a business. The paradigm provides a strategy for operation expansion through FDI. The goal is to determine if a particular approach provides greater overall value than other available national or international choices for the production of goods or services.

Since businesses seek the most cost-effective options while still maintaining quality, they may use the eclectic paradigm to evaluate any scenario which exhibits potential.

Three Key Factors of the Eclectic Paradigm

For FDI to be beneficial, the following advantages must be evident:

The first consideration, ownership advantages , include proprietary information and various ownership rights of a company. These may consist of branding, copyright, trademark or patent rights, plus the use and management of internally-available skills. Ownership advantages are typically considered to be intangible. They include that which gives a competitive advantage, such as a reputation for reliability.

Location advantage is the second necessary good. Companies must assess whether there is a comparative advantage to performing specific functions within a particular nation. Often fixed in nature, these considerations apply to the availability and costs of resources, when functioning in one location compared to another. Location advantage can refer to natural or created resources, but either way, they are generally immobile, requiring a partnership with a foreign investor in that location to be utilized to full advantage.

Finally, internalization advantages , signal when it is better for an organization to produce a particular product in-house, versus contracting with a third-party. At times, it may be more cost-effective for an organization to operate from a different market location while they keep doing the work in-house. If the business decides to outsource  the production, it may require negotiating partnerships with local producers. However, taking an outsourcing route only makes financial sense if the contracting company can meet the organization’s needs and quality standards at a lower cost. Perhaps the foreign company can also offer a greater degree of local market knowledge, or even more skilled employees who can make a better product.

Real World Example

According to Research Methodology, an independent research and analyst firm, the eclectic paradigm were applied by Shanghai Vision Technology Company , in its decision to export its 3D printers and other innovative tech offerings. While their choice strongly considered the disadvantage of higher tariffs and transportation costs, their internationalization strategy ultimately allowed them to flourish in new markets.

• Terms of Service
• Editorial Policy
• Privacy Policy

The OLI Paradigm as a comprehensive model of FDI determinants: a sub-national approach

PurposeThis study aims to advance a sub-national perspective within the OLI Paradigm by analyzing how and to what extent the Eclectic Paradigm can serve as a general model to capture region-specific aspects of the location determinants of FDI, encompassing institutional effects that extend beyond the quality of institutions.Design/methodology/approachThe authors conducted a systematic literature review of 41 selected papers published between 1990 and 2019. Using inductive content analysis, they investigated the theoretical choices used to support analyses of the effects of institutional factors on MNEs' location decisions at the sub-national level.FindingsIt was found that, when changing from the national to the sub-national level of analysis, there is no need to change the main assumptions used in the literature, although a different perspective must be adopted. The Eclectic Paradigm permeates most of the studies revised and can serve as a general model to capture the sub-national perspective. It offers a foundation for new perspectives on the dynamics of institutional and political factors and their effects on location strategies and determinants at the sub-national level. Adopting the OLI Paradigm with a sub-national approach could widen the IB literature's prevailing focus on traditional economic factors and institutional quality.Research limitations/implicationsThe authors contribute to extant International Business literature Their paper enhances the literature on FDI location determinants by providing a more specific approach to development of a sub-national perspective within the OLI Paradigm, extending the institutional effects to capture more region-specific factors influencing the location of FDI. Study limitations are related to our analytical focus on the location dimension, excluding motives for FDI or firm-level location strategies. Rather than limiting analysis to quantitative studies, future research that includes qualitative studies and also covers the other dimensions of the OLI Paradigm could open additional new research avenues for advancing the sub-national perspective within the field of IB.Practical implicationsThe authors’ main findings suggest that MNEs' location strategies should include a sub-national perspective, which means that firms need to assess different levels of the location and understand their interaction with nationwide constraints and limitations, as it may affect firms' ability to effectively conduct their value-adding activities. They also contribute elements that can support sub-national governments' actions and policies aiming to enhance locational advantages to attract and retain FDI.Originality/valueThis review specifically analyzes the location determinants of FDI at the sub-national level, in studies published in a broad set of journals, from a variety of fields, prioritizing articles that investigate sub-national institutional determinants. The authors derive implications for the International Business literature and propose that the sub-national dimension should be incorporated into the Eclectic paradigm in order to better understand the influence of institutional sub-national determinants.

• Related Documents

New institutional economics: contributions to international business studies

Purpose The purpose of this paper is to analyze the main contributions of the new institutional economics to the international business (IB) arena. It also intends to show how the NIE is being incorporated to both eclectic paradigm and Uppsala school’s view, and how it is modifying them. Design/methodology/approach A range of IB articles, which contained an institutional view and also discussed the eclectic paradigm or the Uppsala school, provides the background to build a framework. Findings This paper proposes a framework showing the impact of the institutional variables on the internationalization of firms, by addressing both the OLI paradigm and Uppsala school. It also concludes that the institutional theory has been a point of intersection between the OLI paradigm and Uppsala school, since both have been renewed to understand the transaction costs borne by the firms in their international learning process and in the search for less asymmetrical information. Research limitations/implications This paper provided a brief discussion about the institutional components. Practical implications This study is a useful source of information for those who want to discuss the institutional impact in the IB arena and emerging markets. Originality/value This paper summarizes how the OLI paradigm and Uppsala school encompassed the institutional variables. It also presents a framework that allows new study possibilities since the understanding of the influence of institutional variables on the international movements of firms is still cloudy.

How outward investment from emerging markets affects economic development at home: using the eclectic paradigm to synthesize two IB literatures

Purpose This paper aims to use the eclectic paradigm as a broad organizing framework to bring together two somewhat parallel international business (IB) literatures, one on the development effects of multinational enterprise activity and the other on the internationalization of emerging market multinationals (EMNEs). The author does so to better understand how outward foreign investment shapes economic development in firms’ home countries. Design/methodology/approach Considering that the characteristics of foreign investment by EMNEs likely differ from that of their developed economy counterparts and that such characteristics may have unique development consequences, the author revisits one of IB’s overarching theories to rethink how ownership, location and internalization advantages take shape and stimulate diverse development outcomes. Findings My narrative review and conceptual analysis indicate that the eclectic paradigm is a valuable framework that can be used to shed light on underexplored phenomena and thereby inform important policy debates. The analysis suggests that unique characteristics of EMNE investment simultaneously have positive and negative development consequences in their home countries. Practical implications The author sets out a research agenda that revolves around six propositions that separately relate one of these three distinct characteristics of EMNE investment to two development outcomes, namely, spillovers and direct effects on home-country employment. My propositions suggest that important policy dilemmas potentially apply, in that each of the three characteristics positively affects one of the aspects of development, but negatively the other. Originality/value My research agenda presents international business scholars with new opportunities to build on a history of policymaking impact, now geared toward resolving society’s grand challenge of underdevelopment.

Re-examining regional borders and the multinational enterprise

Purpose – The purpose of this paper is to critically assess the theoretical underpinnings and extant progress of the research on regional multi-national enterprises (MNEs) and offer a blueprint for future research by re-conceptualizing how (regional) boundaries relate to the international diversification of MNEs. Design/methodology/approach – The paper integrates key insights from the theory of the regional MNE and economic geography to re-orient the treatment of regional borders within international business (IB) literature. Findings – The paper suggests that the (L) component within the ownership location and internalization (OLI) paradigm should be disaggregated into continuous “distance effects” and discrete “border effects”. Within this rubric, regional borders represent discrete border effects that generate discontinuities that are permeable, fluid and firm specific. Such reconceptualization opens up avenues for future research and more tightly integrates the research on regional MNEs with other research streams. Research limitations/implications – IB scholars need to make concerted effort to think of regions as one among several parameters in studying the strategy and structure of MNEs. A stronger focus on internal processes and mechanisms elucidating the main drivers of MNEs strategies is needed. Originality/value – The paper offers innovative ways in which future research can advance the study of how regions matter in the internationalization strategy of MNEs.

Crisis or opportunity: Marks and Spencer’s tryst with Indian retail

Research methodology The case is written using secondary data sources (namely, research documents, press information, journal articles and published interviews). Publicly declared company information has further been leveraged to augment case facts. All information sources have been duly acknowledged in the reference section. Case overview/synopsis The case is written in the backdrop of COVID-19 pandemic and its effect on the Indian retail industry, revolving around scenarios in which a multinational retailer has to decide on its long- and short-term strategy in such an economic crisis. The case story has been developed around Marks and Spencer’s retail venture in the Indian market. With the COVID-19 pandemic impacting business at various levels, with countries moving to lock down and economies shrinking to recessionary levels, one of the worst affected sectors is retail. The teaching case builds upon Mark and Spencer’s initial decision of not entering and extending its food/grocery business in India. While it remained a dominant player in Indian fashion retail for almost two decades, it needs to re-think its decision of entering food retail owing to a pandemic situation affecting its offline sales/store footfall and increasing competition from global fashion brands such as Zara and H&M that had flooded the Indian fashion retail sector. The case provides a context for students to perform environmental factor and competitor analysis for a sector, with special focus on decision making in a changing crisis scenario. Complexity academic level This case could be used in undergraduate and MBA classroom programme, across subjects such as retail management, marketing management, international business, international business environment and strategic business management. This case fits while discussing topics such as business environmental factors, competitor analysis, decision-making under crisis, market entry decision, omnichannel retail strategy, consumer behaviour and brand management.

The prevalence of ill-treatment and bullying at work in Ireland

PurposeThe purpose of this paper is to examine the prevalence of ill-treatment and bullying experienced by Irish workers and to explore individual and organisational predictors. The most recent national figures available are specific to bullying and predate the economic recession; therefore, this study is timely and investigates a broader range of negative behaviours.Design/methodology/approachA questionnaire survey study on a national probability sample of Irish employees was conducted (N = 1,764). The study design replicated the methodology employed in the British workplace behaviour study.FindingsThe results showed that 43% of Irish workers had experienced ill-treatment at work over the past two years, with 9% meeting the criteria for experiencing workplace bullying. A number of individual and organisational factors were found to be significantly associated with the experience of ill-treatment at work.Research limitations/implicationsThis study provides national-level data on workplace ill-treatment and bullying that are directly comparable to British study findings.Practical implicationsThe findings indicate that a significant number of Irish workers experience ill-treatment at work, and that workplace bullying does not appear to have decreased since the last national study was conducted in Ireland.Social implicationsThis study is of use to the Irish regulator and persons responsible for managing workplace bullying cases, as it identifies high-risk work situations and contributing individual factors.Originality/valueThis study provides national Irish data on workplace behaviour and ill-treatment following a severe economic recession.

Middle East luxury retail sector – opportunities or uncertainties in the future?

Subject area The subject area is international business and global operations. Study level/applicability The study includes BSc, MSc and MBA students and management trainees who are interested in learning how an industry can be assessed to make a decision on market entry/expansion. Even senior management teams could be targeted in executive education programs, as this case provides a detailed procedure and methodology that is also used by companies (multinational corporations and small- and medium-sized enterprises) to develop strategies on corporate and functional levels. Case overview A group of five senior executive teams of different Swiss luxury and lifestyle companies wanted to enter the Middle East market. To figure out the optimal market entry and operating strategies, the senior executive team approached the Head of the Swiss Business Hub Middle East of Switzerland Global Enterprise, Thomas Meier, in December 2012. Although being marked with great potential and an over-proportional growth, the Middle Eastern luxury market contained impediments that international firms had to take into consideration. Therefore, Thomas had to analyze the future outlook for this segment of the Middle East retail sector to develop potential strategies for the five different Swiss luxury and lifestyle companies to potentially operate successfully in the Middle East luxury and lifestyle market. Expected learning outcomes The study identifies barriers and operations challenges especially for Swiss and other foreign luxury and lifestyle retailers in the Middle East, understands the future (2017) institutional environment of the luxury and lifestyle retail sector in the Middle East and applies the institutions-resources matrix in the context of a Swiss company to evaluate the uncertainties prevailing in the Middle East luxury and lifestyle retail sector. It helps in turning insights about future developments in an industry (segment) into consequences for the corporate and functional strategies of a company. Supplementary materials Teaching notes are available for educators only. Please contact your library to gain login details or e-mail [email protected] to request teaching notes. Subject code CSS 5: International Business.

Social capital and national innovation system: a cross-country analysis

Purpose – In the last two decades, researchers have paid much attention to the role of cultural values on economic and social development. In particular, the crucial role of different aspects of culture on the development of innovation has been stressed in the literature. Consequently, it is vital to understand how social capital, as a core cultural value, affects the innovation process and the innovative performance at the national level. However, to date, the impact of different dimensions of social capital and innovation has not been properly portrayed or explained. Thus, the purpose of this paper is to investigate the influence of four different dimensions of social capital (institutional and interpersonal, associational life and norms) on two of the main functions of national innovation system (NIS) (entrepreneurship and knowledge creation) based on over 50,000 observations in 34 countries. Design/methodology/approach – In this regard, national-level data from the World Values Survey database was employed to quantify social capital. Entrepreneurship is, in turn, assumed to consist of three sub-indexes and 14 indicators based on the Global Entrepreneurship Index. Knowledge creation is also measured through US Patent Office applications. Also, exploratory factor analysis and structural equation modeling approach were used to build the measurement model and investigate the impact that each factor of social capital had on entrepreneurship and knowledge application, respectively. Measurement and structural models were built and their reliability and validity were tested using various fit indices. Research findings suggest the strong positive effect of institutional trust and networking on entrepreneurship. Also, interpersonal trust and networks were shown to have high influence on knowledge development at the national level. Norms appear to have naïve to medium negative effects on both functions. Findings – Research findings suggest the strong positive effect of institutional trust and networking on entrepreneurship. Also, interpersonal trust and networks were shown to have high influence on knowledge development at the national level. Norms appear to have naïve to medium negative effects on both functions. Originality/value – However, to date, the impact of different dimensions of social capital and innovation has not been properly portrayed or explained.

Support for female international business travellers in dual-career families

Purpose The purpose of this paper is to evaluate experiences of female international business travellers living in dual-career families (DCFs) who also have childcare obligations. In particular, the paper explores in which way different sources of support – specifically organizational support – are perceived as important and are available to the women under research. Because of the women’s regular absences due to business trips and the fulfilment of their family role, challenges regarding childcare or household responsibilities may occur. Consequently and also according to social support theory, different types of support may be necessary to organize family and international career, and effectively perform in both environments. Design/methodology/approach This paper examines the experiences and thoughts of these women with special regards on support issues. Data are collected from 51 semi-structured interviews with internationally travelling women in DCF situations originating from seven Western and non-Western countries. The interviews are analysed applying template analysis. Findings Results show that, across countries, support is mainly derived from within family. Regarding institutional or organizational support, however, the reported expectations and actually offered activities differ because of local institutional and cultural variations. The examined women value different forms of organizational support, but do not necessarily expect it. Originality/value This paper represents the first exploratory examinations of various forms of support for female international business travellers in DCFs suggested by social support theory. It includes a culturally diverse sample and contributes to cross-cultural career research.

From Asia to Africa: the international expansion of Hon Chuan enterprise

Subject area International Expansion; Emerging Markets; Corporate Strategy; Strategic Management. Study level/applicability Senior undergraduate; MBA; EMBA. Case overview This case focuses on the international expansion of Hon Chuan Enterprise, a beverage packaging and filling company headquartered in Taiwan. The company has set foot in Africa after its development in mainland China and Southeast Asia. Its 41st factory has just started production in Mozambique, Africa. The African base may help the company reach the turnover milestone of NT$20bn (approximately US$640m) in the next year. This NT$20bn turnover has been a target every year since 2013, but they have so far failed to reach it. As an original equipment manufacturer (OEM) in beverage packaging and filling, Hon Chuan to some extent relies on customers that own brands. After losing a key customer in mainland China, the company has experienced a three-year slump that forced the company’s president, Hish-Chung Tsao, to modify his strategy. Africa was the new battlefield bearing his ambition. His intention was not just to add another manufacturing base, but to develop its own beverage brands as an OBM. Yet, how could this be achieved in Africa? It would be a new journey full of challenges. Africa was more complex than other markets. The company’s first factory there had just been established, and its future was still unknown. Expected learning outcomes This case is appropriate for courses in international business, emerging markets, corporate strategy and marketing management. After studying the case, students should be able to understand international expansion of a manufacturing company in emerging markets; understand several key emerging markets of the world and learn what CAGE distances are; identify Hon Chuan’s success factors, challenges and necessary capabilities for future development and then comprehend why it is important to upgrade from OEM to OBM; and learn how to develop beverage brands in emerging markets. Supplementary materials Teaching Notes are available for educators only. Please contact your library to gain login details or email [email protected] to request teaching notes. Subject code CSS 5: International Business.

Exploring the importance of cultural collectivism on the efficacy of lean practices

Purpose – The purpose of this paper is to assess the influence of cultural collectivism on the efficacy of lean practices. Furthermore, this study assesses whether or not potential cultural disadvantages related to the level of individualism at the national level can be compensated for at the organisational culture level. Design/methodology/approach – Hofstede’s cultural dimension of individualism is used to test whether practicing a collectivistic culture at the organisational level can fully compensate for the potential disadvantages of being geographically situated in an individualistic culture when practicing lean manufacturing. Findings – Results suggest that cultural collectivism at the national and organisational level have a significant impact on the efficacy of lean practices. Furthermore, the negative impact of being situated in an individualistic country cannot be fully compensated for through practicing a collectivistic organisational culture when practicing lean. Originality/value – This study represents a comprehensive attempt to simultaneously assess the collectivism cultural components of lean practices at the national as well as at the organisational level.

Export Citation Format

Share document.

• DOI: 10.18843/IJCMS/V8I3/07
• Corpus ID: 56392721

Understanding Dunning’s Oli Paradigm

• J. Sharmiladevi
• Published 1 September 2017
• Business, Economics

8 Citations

Strategic view of internationalization process, a case study, determinants of foreign direct investment in thailand: chinese investors in chiang mai, thailand.

• Highly Influenced

Competitive advantage and internationalization of a circular economy model in apparel multinationals

The international scientific and practical conference "contemporary issues of economic development of russia: challenges and opportunities" foreign investor’s strategies in the context of the information and innovation paradigm, tax incentives, ease of doing business and inflows of fdi in africa: does governance matter, oversea investment inflows and trade policies in sub-sahara african countries, time and frequency domain causality testing: the causal linkage between fdi and economic risk for the case of turkey, green energy for belt and road initiative: economic aspects today and in the future, 33 references, reappraising the eclectic paradigm in an age of alliance capitalism, the determinants of international production.

• Highly Influential
• 11 Excerpts

A Critical Assessment of the Eclectic Theory of the Multinational Enterprise

Incorporating trade into the investment development path: a case study of korea and taiwan, the eclectic (oli) paradigm of international production: past, present and future, multinational enterprise and economic analysis: preface, the product cycle hypothesis in a new international environment, direct foreign investment : a japanese model of multinational business operations, internalization theory and corporate international finance, relational assets, networks and international business activity, related papers.

Showing 1 through 3 of 0 Related Papers

Browse Econ Literature

• Working papers
• Software components
• Book chapters
• JEL classification

More features

• Subscribe to new research

RePEc Biblio

Author registration.

• Economics Virtual Seminar Calendar NEW!

Some searches may not work properly. We apologize for the inconvenience.

Understanding Dunning’s OLI Paradigm

• Author & abstract
• 11 References
• Most related
• Related works & more

Corrections

(Assistant Professor, Symbiosis Centre for Management Studies, Symbiosis International University, Pune, India.)

Suggested Citation

Download full text from publisher, references listed on ideas.

Follow serials, authors, keywords & more

Public profiles for Economics researchers

Various research rankings in Economics

RePEc Genealogy

Who was a student of whom, using RePEc

Curated articles & papers on economics topics

Upload your paper to be listed on RePEc and IDEAS

New papers by email

Subscribe to new additions to RePEc

EconAcademics

Blog aggregator for economics research

Cases of plagiarism in Economics

About RePEc

Initiative for open bibliographies in Economics

News about RePEc

Questions about IDEAS and RePEc

RePEc volunteers

Participating archives

Publishers indexing in RePEc

Privacy statement

Found an error or omission?

Opportunities to help RePEc

Get papers listed

Have your research listed on RePEc

Open a RePEc archive

Have your institution's/publisher's output listed on RePEc

Get RePEc data

Use data assembled by RePEc

Breadcrumbs Section. Click here to navigate to respective pages.

International Business and the Eclectic Paradigm

DOI link for International Business and the Eclectic Paradigm

Get Citation

The eclectic paradigm has arguably become the dominant theoretical basis in the study of FDI, multinational corporations and internationalisation over the last two decades. The contributions to this volume evaluate the eclectic paradigm in the global economy and its validity as a theoretical basis to understand developments such as economic globali

TABLE OF CONTENTS

Chapter 1 | 20  pages, revisiting the eclectic paradigm: new developments and current issues, chapter 2 | 19  pages, the eclectic (oli) paradigm of international production: past, present and future, chapter 3 | 24  pages, new institutional economics: an organising framework for oli, chapter 4 | 20  pages, fdi and endogenous growth: ib perspectives, chapter 5 | 14  pages, the co-evolutional advantage: strategic management theory and the eclectic paradigm, chapter 6 | 16  pages, from oli to olma: incorporation higher levels of environmental and structural complexity into the eclectic paradigm, chapter 7 | 20  pages, from a theory to a paradigm: examining the eclectic paradigm as a framework in international economics, chapter 8 | 18  pages, managerial beliefs, market contestability and dominant strategic orientation in the eclectic paradigm, chapter 9 | 23  pages, the micro-mechanics of foreign operations' performance: an analysis based on the oli framework, chapter 10 | 17  pages, the eclectic paradigm and the recognition of finance-specific factors, chapter 11 | 29  pages, the challenge of electronic markets for international business theory, chapter 12 | 23  pages, the recent globalisation of the italian banking sector: an interpretation based on the eclectic paradigm, chapter 13 | 19  pages, a critical reflection and some conclusions on oli.

• Privacy Policy
• Terms & Conditions
• Cookie Policy
• Taylor & Francis Online
• Taylor & Francis Group
• Students/Researchers
• Librarians/Institutions

Connect with us

Registered in England & Wales No. 3099067 5 Howick Place | London | SW1P 1WG © 2024 Informa UK Limited

Loading metrics

Open Access

Peer-reviewed

Research Article

pyPAGE: A framework for Addressing biases in gene-set enrichment analysis—A case study on Alzheimer’s disease

Roles Conceptualization, Data curation, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing

Affiliation Faculty of Bioengineering and Bioinformatics, Lomonosov Moscow State University, Moscow, Russia

Roles Software

Affiliation Institute for Neurodegenerative Diseases, University of California San Francisco, California, United States of America

Roles Conceptualization, Writing – review & editing

Affiliations Institute for Neurodegenerative Diseases, University of California San Francisco, California, United States of America, Department of Biochemistry and Biophysics, University of California San Francisco, San Francisco, California, United States of America

Roles Conceptualization, Methodology, Software, Validation, Writing – original draft

* E-mail: [email protected] ; [email protected]

Affiliations Department of Biochemistry and Biophysics, University of California San Francisco, San Francisco, California, United States of America, Department of Urology, University of California San Francisco, San Francisco, California, United States of America, Helen Diller Family Comprehensive Cancer Center, University of California San Francisco, San Francisco, California, United States of America, Bakar Computational Health Sciences Institute, University of California San Francisco, San Francisco, California, United States of America

Roles Conceptualization, Supervision, Writing – original draft, Writing – review & editing

Affiliations Department of Biochemistry and Biophysics, University of California San Francisco, San Francisco, California, United States of America, Department of Urology, University of California San Francisco, San Francisco, California, United States of America, Helen Diller Family Comprehensive Cancer Center, University of California San Francisco, San Francisco, California, United States of America, Bakar Computational Health Sciences Institute, University of California San Francisco, San Francisco, California, United States of America, Arc Institute, Palo Alto, California, United States of America

• Artemy Bakulin, 
• Noam B. Teyssier, 
• Martin Kampmann, 
• Matvei Khoroshkin, 
• Hani Goodarzi

• Published: September 5, 2024
• https://doi.org/10.1371/journal.pcbi.1012346
• Peer Review
• Reader Comments

This is an uncorrected proof.

Inferring the driving regulatory programs from comparative analysis of gene expression data is a cornerstone of systems biology. Many computational frameworks were developed to address this problem, including our iPAGE ( i nformation-theoretic P athway A nalysis of G ene E xpression) toolset that uses information theory to detect non-random patterns of expression associated with given pathways or regulons. Our recent observations, however, indicate that existing approaches are susceptible to the technical biases that are inherent to most real world annotations. To address this, we have extended our information-theoretic framework to account for specific biases and artifacts in biological networks using the concept of conditional information. To showcase pyPAGE, we performed a comprehensive analysis of regulatory perturbations that underlie the molecular etiology of Alzheimer’s disease (AD). pyPAGE successfully recapitulated several known AD-associated gene expression programs. We also discovered several additional regulons whose differential activity is significantly associated with AD. We further explored how these regulators relate to pathological processes in AD through cell-type specific analysis of single cell and spatial gene expression datasets. Our findings showcase the utility of pyPAGE as a precise and reliable biomarker discovery in complex diseases such as Alzheimer’s disease.

Author summary

Biological regulation is governed by a complex network of interactions involving transcription factors, RNA-binding proteins, and microRNAs. To reveal the regulatory programs underlying gene expression modulations, researchers often take advantage of gene-set enrichment analysis, an approach that studies concerted changes in a group of genes rather than observing genes in isolation. Previously, we developed a tool called iPAGE to facilitate this analysis. However, both iPAGE and other similar tools implicitly assume that different genes have uniform gene-set membership. Our recent observations challenge this assumption, revealing that some genes form regulatory networks far more frequently than others. These technical biases and redundancies in gene-set annotations complicate the accurate inference of true regulatory relationships in specific contexts. To overcome these limitations, we introduced pyPAGE, an enhanced method that extends our information-theoretic framework by incorporating conditional mutual information to account for specific biases and artifacts.

We applied pyPAGE to Alzheimer’s disease, a neurodegenerative disorder marked by its complexity and characterized by progressive cognitive decline, memory loss, and behavioral changes. Pathological processes of Alzheimer’s disease involve various cell types and diverse molecular mechanisms, which pose significant challenges for study. In our study we took advantage of our newly developed framework pyPAGE, performing analysis of gene expression changes at both tissue and cell type levels. This way we were able to describe regulation patterns of known regulators of Alzheimer’s as well as several new ones. Additionally, we performed a cohort study of the association between the enrichment of the identified regulons and survival prognosis for patients, showing that increased activity of a subset of RBPs is positively associated with a lifespan. In total the results highlight the utility of pyPAGE and provide a valuable set of biomarkers for Alzheimer’s disease.

Citation: Bakulin A, Teyssier NB, Kampmann M, Khoroshkin M, Goodarzi H (2024) pyPAGE: A framework for Addressing biases in gene-set enrichment analysis—A case study on Alzheimer’s disease. PLoS Comput Biol 20(9): e1012346. https://doi.org/10.1371/journal.pcbi.1012346

Editor: Andrey Rzhetsky, University of Chicago, UNITED STATES OF AMERICA

Received: August 31, 2023; Accepted: July 22, 2024; Published: September 5, 2024

Copyright: © 2024 Bakulin et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: MSBB and ROSMAP data can be found in Synapse database and will not be shared with this paper. This data is individual level and therefore in compliance with Sage Bionetworks policy is under controlled access. To get an access to it a Data Use Certificate should be submitted to the AD knowledge portal with the description of a study plan. Any questions regarding data access can be addressed to the help desk [email protected] . All results of our analysis including processed eCLIP data are available in a supplement. The program code is deposited at https://github.com/goodarzilab/pypage .

Funding: The work was supported by a number of grants for neurodegenerative research received by MK. These include NIH grants R01 AG062359, U54 NS123746, U01 AG072464 and a Chan Zuckerberg Initiative Ben Barres Early Career Acceleration Award. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Over the past two decades, gene expression analysis has emerged as an effective approach for defining cell states at a molecular level. For example, disease signature genes, defined as genes that are up or down-regulated in diseased cells, have long been used as a proxy for pathological cell identity [ 1 , 2 ]. However, even in the early days, it was clear that for complex diseases like cancer, signature genes fail to reproduce in independent studies [ 3 ]. This was not surprising, as it is the collective disruptions in cellular pathways and processes, rather than dysregulated expression of individual genes themselves that underlie cellular pathologies. This understanding gave rise to a suite of computational analyses that perform pathway-level or gene-set-level comparisons between samples. Broadly defined, gene-sets include (i) pathways or complexes, sets of functionally related genes that act together or sequentially, and (ii) regulons, genes that are downstream of a given regulatory factor. The latter provides an opportunity to assess the differential activity of master regulators across conditions [ 4 ]. Regulons are typically annotated using experimental data (such as ChIP-seq [ 5 ] and CLIP-seq [ 6 ]) or inferred using sequence binding preferences of each regulator (AlignACE [ 7 , 8 ], FIRE [ 9 ], DeepBind [ 10 ], DeepSEA [ 11 ], DanQ [ 12 ]). We have previously shown that application of this methodology to large scale gene expression datasets reveals the underlying regulatory programs that are key to the emergence or maintenance of a given cellular state [ 1 ]. However, as we will discuss here, additional challenges needed to be addressed to further improve our ability to detect perturbations in regulatory programs.

There are myriad tools implementing pathway level analysis; Garcia-Campos et al [ 13 ] categorize these tools into two main categories: over-representation analysis (ORA) methods and Functional Class Scoring (FCS) methods. ORA relies on the concept that the top-scoring genes should carry the highest weight; therefore, the ORA methods operate under the null hypothesis that if a number of differentially expressed genes is chosen, the fraction of genes that belong to any given pathway does not exceed the fraction expected by chance. In contrast, FCS methods operate under the assumption that pathways might consist of many genes that show heterogeneous yet coordinated changes in expression. Such methods make use of all available measurements and often do not rely on arbitrary thresholding.

In these methodologies, pathway annotation databases, such as Gene Ontology [ 14 , 15 ] and MSigDB [ 3 , 16 ], can be used to represent pathways as gene-sets or binary vectors where a gene can either belong to a given pathway, or not. However, this commonly used representation of pathway membership creates an intrinsic bias for several reasons. First, genes are multifunctional and some genes are far more studied than others and therefore are simultaneously annotated as part of many pathways. Second, the more abundant a given protein, the more often it is detected in various assays (such as affinity purification); therefore, abundant proteins similarly tend to be categorized as members of many complexes and gene-sets. Such biases can have a broad impact on target selection and pathway identification in large-scale datasets. For instance, identifying a protein involved in only a few gene-sets as differentially expressed narrows the range of pathways considered relevant, compared to a similar finding about a protein associated with many gene-sets. To our knowledge, no available pathway-analytical tool explicitly takes such biases into account.

Here, we propose a pathway analytical approach that explicitly models the representation bias present in gene-set annotations. This framework is an extension of our iPAGE package developed more than a decade ago [ 1 ]. In iPAGEv1.0 we used mutual information to directly estimate the dependency between expression values and gene-set annotations. In contrast, pyPAGE relies on calculating conditional mutual information, which allows for explicit consideration of annotation bias as a covariate. Here, we provide a number of benchmarks showcasing the utility of pyPAGE as a robust pathway analytical tool that surpasses its predecessors. We demonstrate that pyPAGE can effectively reveal the regulatory mechanisms underlying biological observations, which is a crucial first step towards a molecular understanding of their origin and maintenance.

The generalizability of pyPAGE allows it to infer regulatory changes at various molecular levels using a variety of different data modalities, including differential gene expression and differential stability estimates inferred from both bulk and single cell data. We took advantage of this to carry out a comprehensive analysis of gene expression programs in Alzheimer’s disease. While AD is currently characterized at molecular, histological, and symptom levels [ 17 ], the regulatory networks, both transcriptional and post-transcriptional, that accompany and likely drive cellular pathogenesis in AD are not fully explored. Leveraging pyPAGE, we have generated novel insights into the role of key master regulators, including transcription factors (TF), RNA binding proteins (RBP), and microRNAs (miRNAs), in shaping the pathological transcriptome in AD. First, we perform an analysis of bulk RNA-sequencing data from Mount Sinai Brain Bank (MSBB) [ 18 ], a consortium that holds a large collection of characterized brain data from both healthy donors and patients with pathological conditions. We focus on the analysis of samples of Brodmann Area 36, since it is one of the most vulnerable brain regions in AD [ 19 ] and we aimed at identifying a core set of deregulated regulons that are inherent to the disease. Next we provide a cell-type specific characterization of differential activity of the identified regulons using two single-cell RNA sequencing datasets of human prefrontal cortex [ 20 , 21 ], followed by description of their spatial regulation patterns inferred from spatial transcriptomics data of middle temporal gyrus [ 22 ]. Though we analyze different brain regions, we show an overlap between their regulatory signatures. Finally, on a different bulk RNA-sequencing dataset of dorsolateral prefrontal cortex we showcase the utility of the identified factors for a stratification of patients based on the activity of these factors inferred using pyPAGE. Using pyPAGE in these analyses provides a significant advantage by effectively accounting for uncertainty around putative regulatory interactions and non-specific and artifactual hits, therefore improving our ability to capture real biological signals in the data.

Gene-set annotations often form scale-free networks

In every annotation of biological regulons, gene-set membership varies significantly among all genes. Typically, this measure of gene-set membership reflects the number of functional interactions a gene engages in. Current analytical tools assume gene-set membership to follow a normal distribution ( Fig 1A ). However, in practice this assumption rarely holds true and gene-set membership instead follows the power law ( Fig 1A ). We have demonstrated this across multiple commonly used gene-set annotations: molecular pathways, transcription factor target genes, RNA-binding protein regulons and miRNAs targets ( S1A–S1D Fig ). In each of these annotations, the degree distribution of the gene-set membership could be closely approximated using the power law function (R 2 > 0.9), which typically peaks at 1 with a long distribution tail. This implies that like most biological networks, gene-set annotations also form scale-free networks.

• PPT PowerPoint slide
• PNG larger image
• TIFF original image

(A) Comparison between the theoretical and empirical degree distributions of gene-set membership in gene-set annotations. The red line represents the curve-fitting of the power law function to the observed distribution. (B) Scatter plots representing major sources of bias in biological annotations. The left panel represents the association between the protein abundance and the number of interactions a gene has in the STRING database. The right panel represents the association between the citation index of a particular gene and its gene-set membership in an annotation of biological pathways. For each association, we also report correlation between values. (C) Characteristics of gene-set membership degree distribution within TF regulon annotations are depicted. The top plot displays the observed distribution with a power law function fitted to it, including reporting the gamma parameter and the R^2 value. The bottom plot illustrates the deviation of the observed distribution from the expected power law.

https://doi.org/10.1371/journal.pcbi.1012346.g001

Further analysis of these gene-set membership distributions revealed that in many cases the shape of the right tails significantly deviates from the power law ( Fig 1C ). Such deviations indicate a bias in the underlying annotations. We posit that this bias arises from the experimental and analytical techniques that are used to compile them. For example, we observed that the more a given gene is studied the more likely it is to have multiple entries in a knowledge-based annotation ( Fig 1B , right panel ). Another source of bias may be technical limitations including low sensitivity of detecting weak functional interactions. Consistently, we have observed that, in the human interactome, the number of physical interactions for a given protein correlates with its protein abundance ( Fig 1B , left panel ). Overall, these findings indicate that there are important structural biases in the commonly used gene-set annotations. As we will demonstrate below, these inherent biases, which are overlooked in most gene-set enrichment analyses, have an impact on the insights that can be gleaned from the underlying data using current methodologies.

pyPAGE overcomes structural annotational biases in gene-set enrichment analysis

Current gene-set enrichment analysis tools assume that every gene in a regulon is equally informative of regulon activity. However, as we demonstrated above, gene-set membership significantly varies between genes. As a consequence, differential expression of genes with specific functions provides stronger evidence for deregulation of its corresponding gene-sets than changes in expression of a more promiscuous gene. To address these biases and imbalances, we developed pyPAGE, an improved gene-set enrichment analysis method. pyPAGE is an extension of our previously published iPAGE framework [ 1 ] and can be used to analyze various gene-set annotations such as Gene Ontology, MSigDB, and regulons of TFs, RBPs and microRNAs. To correct for imbalances in gene-set membership, pyPAGE uses the concept of conditional mutual information [ 23 ] and measures the dependency between input data (e.g. differential gene expression) and a specific gene-set while taking into account the total number of gene-sets every given gene belongs to. The statistical significance of this dependency is then determined using a permutation test. Like its predecessor, pyPAGE detects any non-random patterns of enrichment or depletion of target genes and even non-monotonic complex relationships are captured (e.g. dual regulators). It also does not depend on the particular distribution of the input data and can be applied to both discrete (e.g. cluster indices) and continuous (e.g. log-fold changes) inputs.

pyPAGE allows for the comprehensive analysis of gene expression changes at multiple levels, including biological pathways as well as transcriptional and post-transcriptional regulons. To standardize the analysis of biological datasets using pyPAGE, we propose the workflow described in Fig 2A . We begin with initial preprocessing of RNA-seq data to obtain estimates of transcript abundance in each sample. Then, for each gene, we compute an estimate of its log fold change and a corresponding p-value using any of the available tools, such as DESeq2 [ 24 ], limma [ 25 ] or edgeR [ 26 ]. These values form the basis for an estimate of differential gene expression as demonstrated by Formula 3 . Next, we apply pyPAGE to identify biological pathways and transcriptional factor regulons that are significantly associated with the differential gene expression. This approach can be broadly used for both bulk and single cell RNA-seq data, which in the latter case would allow better capture of cell-type specific patterns of regulation. Since miRNAs and RBPs impact gene expression by influencing RNA decay, differential RNA stability measurements provide a superior source of signal for detecting significant post-transcriptional regulons. To incorporate this information, we propose using our recently published tool REMBRANDTS, which infers changes in transcript stability by comparing the numbers of intronic and exonic reads [ 27 ]. These measurements can then be used to estimate the magnitude of stability changes between two groups and its significance, from which differential stability can be calculated in a similar way to differential gene expression. In addition to the comparative analysis between groups of samples, pyPAGE supports analysis of regulons in each sample individually using expression or stability estimates that are normalized across samples.

(A) Schematic of the pipeline we propose for the analysis of bulk RNA-seq data using pyPAGE. The pipeline starts with preprocessing of RNA-seq data and then diverges into two branches: one for the analysis of transcriptional regulation and the other for the analysis of post-transcriptional regulation. (B) Precision-recall curves demonstrating the performance of pyPAGE and benchmarking it against iPAGE and fgsea. The analysis was made in 4 simulated scenarios with and without added biases and with or without dual regulation patterns. As a general metric of performance we report PR-AUC score, also cross glyphs mark the performance at p-value threshold equal to 0.01. (C) Graphical representation of pyPAGE’s robustness to variations in input data quality. The analysis incorporates two distinct curves illustrating the effects of: 1) subsampling the data from 5% to 100% in increments of 5%, and 2) adjusting the parameter that dictates the fraction of deregulated genes within each regulon (note that the default value for this parameter is 0.5 which explains divergence of two curves at 1.0).

https://doi.org/10.1371/journal.pcbi.1012346.g002

pyPAGE outperforms existing tools on simulated datasets

To assess the ability of pyPAGE to account for biases while maintaining a high baseline performance relative to its predecessors, we used several simulation-based validation strategies. By controlling the "ground truth" signal in simulated data and setting a priori expectations for gene-set modulations, we were able to measure the sensitivity of pyPAGE and compare its performance to other similar tools.

The main objectives of these simulations were (i) to demonstrate that implicit biases in gene-set annotations can negatively impact the quality of gene-set enrichment analysis, and (ii) to assess whether pyPAGE effectively addresses this issue and robustly identifies deregulation of gene-sets from biased annotations. To simulate pathway annotations, we randomly selected subsets of varying sizes from a set of genes. We introduced artifacts by enhancing the likelihood that certain genes would be included across all pathway annotations. Specifically, we designated a set of ’biased genes’ and added a random sample from this group to each simulated pathway, as detailed in the Methods section. Our secondary goal was to demonstrate that pyPAGE can detect any non-random patterns in data, both monotonic and non-monotonic. Though activity of most factors is associated with either upregulation or downregulation of their target regulons, there are nevertheless factors that function as dual regulators simultaneously exerting both effects. Therefore, we simulated two types of datasets with monotonic and dual regulation patterns.

We assessed iPAGE [ 1 ] and fgsea [ 28 ] as comparators to pyPAGE. The performance of each method was measured by its ability to recover the artificially perturbed gene-sets in each simulated dataset. We employed four simulation scenarios to test how well each algorithm handles bias in annotations and captures monotonic and non-monotonic regulatory patterns. In each case, we report relevant performance metrics ( S2 Fig ). However, since the number of unperturbed gene-sets in the simulation is much greater than the number of perturbed gene-sets, we specifically focus on the precision-recall curves ( Fig 2B ). In these benchmarks, pyPAGE and iPAGE performed equally well predicting deregulation of gene-sets from unbiased annotations both with monotonic and non-monotonic signals (PR-AUC>0.90). However, when using biased annotations, performance of iPAGE decreased dramatically by more than 30% as indicated by changes in PR-AUC score. Performance of pyPAGE, on the other hand, remained largely unchanged. Fgsea performed worse than pyPAGE and iPAGE using the same simulated datasets across all tasks. Firstly, it failed to accept several gene-sets as input and could not analyze their deregulation because of imbalances in their input distributions. Secondly, it largely failed to capture non-monotonic and bimodal signals, which led to a 35% decrease in PR-AUC score as compared to the case with monotonic signals. Finally, annotational biases led to reductions in its PR-AUC by 18% and 30% for monotonic and non-monotonic signals, respectively. These results demonstrate that based on the criteria formulated by our two validation objectives, pyPAGE outperforms other tools in being robust to biases in gene-set annotations and identifying complex patterns in data.

pyPAGE is both robust and sensitive in detecting dysregulated gene-sets

To further assess the robustness and sensitivity of pyPAGE, we implemented several additional simulation strategies. Our goal was to demonstrate that the identified gene-sets remained consistent across independent runs, even as signal-to-noise was reduced. To suppress the signal in the input data, we employed two approaches. First, to measure the sensitivity of pyPAGE to the strength of the input signal, we directly controlled the magnitude of the simulated gene expression changes by tuning it as a hyperparameter. In the second approach we simulated new datasets by randomly selecting and changing a subset of genes in the target gene-set to evaluate pyPAGE’s robustness to incomplete data or imperfect gene-set annotations [ 9 ]. pyPAGE proved to be highly robust to the choice of simulation parameters controlling signal expressiveness within a wide range ( Fig 2C ), which is showcased by its ability to capture more than 80% of the target gene-sets even after an 80% decline in the strength of a signal controlled by the simulation parameters. Similarly, even after sub-sampling the genes to as low as 20% of the original counts, pyPAGE was still able to recover more than 80% of the perturbed target gene-sets. These results suggest a strong potential for pyPAGE to be highly useful on single-cell RNA-seq datasets, where substantially fewer genes are captured relative to bulk data.

Comprehensive analysis of regulatory perturbations in Alzheimer’s disease

As we showed above, pyPAGE provides a sensitive and robust approach for identifying the regulatory programs that drive changes in gene expression. Extending our work to real-world data, we sought to use pyPAGE to study regulatory perturbations in Alzheimer’s disease (AD). This highly complex disease involves dysregulation of many cellular pathways and processes. The recent explosion in the amount of transcriptomic data from AD patients presents an opportunity to systematically investigate the transcriptional and post-transcriptional regulatory programs mediating AD–a first step toward potential new therapeutic targets.

AD primarily causes the atrophy of the large cortex, subcortex and a number of other areas of the brain [ 29 ]. Pathologically, it is characterized by neuronal loss and destruction of synaptic connections, which are linked to the accumulation of amyloid beta and tau protein. Previous research also points to a variety of other pathological mechanisms that may be involved in AD, such as immune dyshomeostasis [ 30 ], oxidative stress [ 31 ] and metal imbalance [ 32 ]. A handful of studies have sought to identify master regulators of these changes based on the regulators’ differential gene expression [ 33 – 35 ]. However, since the activity of proteins is often controlled at the post-translational level [ 36 ], it cannot be reliably inferred from their mRNA expression levels alone. Therefore, we sought to instead infer their differential activity from changes in the expression of their downstream target genes. In our earlier work, we demonstrated that an information-theoretic approach can be effectively employed to capture regulatory programs that underlie complex human diseases [ 1 ]. pyPAGE provides the opportunity to infer associations between AD pathology and deregulation of TFs, miRNAs, and RBPs capitalizing on both known (sourced from curated databases) and putative (derived from high-throughput assays) interaction datasets ( Table 1 ).

https://doi.org/10.1371/journal.pcbi.1012346.t001

Using publicly available RNA-seq data from the MSBB study [ 18 ], we performed a comparison of samples from the BM36 brain region between individuals with an AD diagnosis and without, while controlling for biological variables such as sex. To capture differences in transcriptional control, we used our previously curated annotation of cis -regulatory elements and transcription factor binding sites [ 37 ]. Similarly, post-transcriptional regulatory changes were predicted based on the combined annotation of known RBP binding sites from various sources (see Methods ) and miRNA recognition elements [ 38 , 39 ]. To ensure that the identified regulatory programs reflect AD pathogenesis and not changes in cellular composition artificially read out in bulk data, we also analyzed their cell-type specific activation patterns using single cell measurements, as well as their clinical relevance in predicting long-term survival of AD patients.

Transcriptional regulatory modules associated with Alzheimer’s disease

In our analysis of transcriptional deregulation in AD, we observed significant disease associations for 15 transcription factors (TFs) ( Fig 3A ) most of which are regulators of neuronal functions ( Fig 3E ). We also demonstrate that changes in expression of these TF regulons occur in both female and male donors ( S3A Fig ) and in each age group ( S3B Fig ). For a number of these regulons, we also observed concomitant changes in the expression levels of their associated TFs as well ( Fig 3B ). While the correlated expression of a TF and its regulon strengthens the evidence for the observed associations, as mentioned earlier, differential expression of a TF is only one possible way that it may acquire differential activity. For example, our analysis correctly identified two well-known transcriptional regulators of AD, KDM5A [ 40 ] and ATF4 [ 41 ]–but only the mRNA expression level of KDM5A showed a high correlation with the mean expression of its regulon (R 2 = 0.85) ( Fig 3C ) while ATF4 did not (R 2 = 0.40) ( Fig 3D ). Indeed, ATF4 levels are known to be controlled at the level of translation [ 42 ].

(A) Regulons of TFs differentially expressed between AD and non-AD samples discovered by pyPAGE. In this representation the rows correspond to TFs and columns to gene bins of equal size ordered by differential expression, the cells are colored according to the enrichment of genes from regulons in a corresponding bin. The leftmost column of the heatmap depicts the differential expression of the regulator itself. (B) The barplot representing Pearson correlations between the expression of TFs and of their regulons, as measured by median TPM of its members. Asterix indicated significant correlation (p-value<0.05). (C) The scatter plot demonstrating association between the expression of the well-known AD regulator KDM5A with the expression of its regulon. (D) The association between the expression of another AD regulator ATF4 with the expression of its regulon. (E) Biological roles of the identified TFs inferred based on the functions of the genes controlled by these TFs. In these heatmap colored cells correspond to TFs whose regulons are significantly (p-value<0.05) enriched with genes from a corresponding biological pathway based on PantherDB. (F) Plot showcasing how robust are predictions of three different methods to subsampling of expression data. To measure consistency of the predictions we computed intersection over union (IoU) of the method’s output with and without subsampling of genes.

https://doi.org/10.1371/journal.pcbi.1012346.g003

These transcriptional signatures are robustly recovered by pyPAGE even when part of the input expression data is missing due to the subsampling of genes ( Fig 3F ). By conditioning on the regulon membership of each gene, pyPAGE avoids reporting spurious non-robust associations and outperforms iPAGE by reporting a more consistent set of genes with different fractions of the input (as indicated by higher intersection over union). Fgsea similarly shows high consistency of its predictions across runs with ablated data. However, in contrast to pyPAGE, which identifies deregulation of only 3% of all TF regulons, Fgsea detects significant associations with over half (53%) of all TFs, which is highly unlikely to be biologically meaningful. Indeed, our previous studies of the pathological hallmarks of AD, specifically tau accumulation [ 43 ] and microglia activation [ 40 ], revealed that fewer than 10 out of the top 100 genes associated with these changes were transcription factors, underscoring that AD pathology is driven by the modulation of a limited number of key factors. This suggests that many of the Fgsea hits could be false positives, in line with our observations when benchmarking against synthetic data ( Fig 2B ). Overall, the analysis highlights that pyPAGE is adept at identifying both robust and specific signals within AD datasets.

Consistent with their association with AD pathology, many of the identified TFs are key regulators of brain development and homeostasis. Systematic analysis of the functions represented among the regulons of these identified TFs (see Methods ), revealed that 11 out of 15 TFs regulate molecular programs related to neurogenesis ( Figs 3E and S4 and S1 Table ). For example, ATF4 maintains excitatory properties of neurons [ 44 ], whereas the generation of oligodendrocytes [ 45 ] is controlled by SOX10. While the majority of the identified TFs function as transcriptional activators, two of them, JARID2 and KDM5A, are also involved in epigenetic reprogramming, which is a hallmark of AD pathology [ 46 , 47 ]. Additionally, several of the identified factors control adaptive immune response and protein metabolism, linking these regulons to other key hallmarks of AD: inflammation, amyloid accumulation and abnormal phosphorylation of tau protein.

Cell-type specific patterns of deregulation for AD-associated TF regulons

Since the analysis of bulk data captures a mixture of cell types, it often requires major shifts in gene regulation in order to generate a discernible signal. Moreover, some of the patterns observed in bulk may reflect changes in cell-type abundance rather than bona-fide reprogramming of gene expression. To account for this possibility and improve resolution of our analysis, we applied pyPAGE to the ROSMAP single-cell RNA-seq study [ 20 ], which includes the major brain cell types in samples from both AD patients and healthy donors ( Fig 4A ). For each cell type, we estimated differential gene expression between patients and healthy controls and then applied pyPAGE to predict deregulated TF activity. Since the approach implemented by pyPAGE is independent of any particular distribution of the input data, it was seamlessly extended to the scRNA-seq data domain. The benefits of using pyPAGE over other tools developed for analysis of scRNA-seq data (e.g., Vision [ 48 ], Triangulate [ 49 ], scFAN [ 50 ]) are the same as the ones we enumerated for bulk datasets. As we showed earlier ( Fig 2C ) pyPAGE is robust to data sparsity and it accounts for biases in our annotations. As a result, we expect a reduction in the number of false positives, which makes pyPAGE especially well-suited for regulon exploration of cell-type specific gene expression patterns.

(A) Cells from the analyzed ROSMAP dataset represented on a force-directed graph embedding. The clusters are colored according to cell-types: excitatory neurons (Ex), inhibitory neurons (In), astrocytes (Ast), oligodendrocytes (Oli), oligodendrocyte progenitor cells (Opc), microglia (Mic), endothelial cells (End), pericytes (Per). ( B) The same cell-type clusters colored according to differential activity of SOX10 between cells from non-AD and AD samples estimated using pyPAGE. The magnitude of the regulation pattern was calculated as scaled conditional mutual information multiplied by the factor representing the direction of deregulation. ( C) Summary of the cell-type specific deregulation patterns of the TFs identified in the analysis of the bulk data. Heatmap cells with significant associations (p-value<0.05) are framed. The regulation is calculated as the normalized conditional mutual information of the relationship multiplied by the sign of the log fold change. (D) Heatmap representations of concordant expression changes in expression of TF target genes in inhibitory neurons and oligodendrocytes. Here rows correspond to TFs and columns to gene bins of equal size ordered by differential expression, the cells are colored according to the enrichment of genes from regulons in a corresponding bin. (E) This heatmap summarizes deregulation patterns in various cortical layers of TFs that we previously identified in the analysis of bulk data. Heatmap cells with significant associations (p-value<0.05) are framed. Regulation pattern is estimated as normalized conditional mutual information of the association multiplied by the sign of log fold change.

https://doi.org/10.1371/journal.pcbi.1012346.g004

As mentioned above, the primary goal of integrating single-cell data into our AD analysis workflow was to measure contribution of the identified regulons to the gene expression of individual cell types. As expected, all of the 15 TF regulons we identified from bulk data showed significant deregulation in at least one cell type. Strikingly, 6 of them showed significant associations with gene expression across most of the major cell types, indicating their diverse roles in brain homeostasis. For example, ATF4 is known to act in a cell type specific manner and depending on the cellular context, it activates both pro-survival and pro-death pathways [ 51 ]. Accordingly, in our analysis, we observed that the ATF4 regulon showed divergent patterns of activity in various cell types. SOX10 being a major regulator of oligodendrocyte survival [ 45 ] and inhibitor of astrocyte differentiation [ 52 ] showed increased activity in the former cell type and decreased activity in the latter cell type( Fig 4B ). Or, similarly, TCF4 which is known to be a major regulator of neurogenesis [ 53 ] showed upregulation of its regulon in excitatory neurons and downregulation in inhibitory neurons. Finally, we observed strong evidence of epigenetic reprogramming, manifested either by activation or repression of chromatin modifiers KDM5A and JARID2 target regulons in all major cell types. Taken together, these findings illustrate cell type specific regulatory programs involved in AD, further highlighting the generalizability of pyPAGE to single cell transcriptomic measurements.

Next, we asked whether there were patterns in the single-cell data that had remained hidden in our analysis of bulk RNA-seq. We analyzed the cell type specific differential regulon activity of factors that were not identified in the bulk analysis ( S5A–S5E Fig and Table 2 ). Intriguingly, inhibitory neurons showed downregulation of factors involved in neuronal differentiation (DMT3A [ 54 ], NFKB1 [ 55 ], KDM5A [ 56 ]) ( Fig 4D , top panel ) and, likewise, in excitatory neurons we observed inhibition of TF regulons that characterize mature neurons (NCOR2 [ 57 ], KDM5A [ 56 ]) ( S5B Fig ). Excitatory neurons also showed activation of regulons controlled by neuroprotective factors (SIX2 [ 58 ], NR1H3 [ 59 ], TCF7 [ 60 ]) ( S5B Fig ). Unlike inhibitory neurons, glial cells showed upregulation of developmental factors and similarly to excitatory neurons activation of cytoprotective factors, such as shown, for example, by activation of regeneration programs in oligodendrocytes (NKX3-1 [ 61 ], NFKB1 [ 62 ], EMX1 [ 63 ], IRF2 [ 64 ]) ( Fig 4D , bottom panel ). Together, these observations support the hypothesis that AD pathology disrupts gene expression programs that give rise to and maintain cellular identity in neurons [ 65 ], while simultaneously triggering cytoprotective response in glia.

https://doi.org/10.1371/journal.pcbi.1012346.t002

Spatial patterns of deregulation for AD-associated TF regulons

Beyond the cell type specificity of gene expression programs, the spatial patterning and physical organization of these cell types and functions within the brain can influence physiology in ways that cannot be gleaned from bulk or single cell RNA-seq data. Therefore, we extended our analysis to spatial transcriptomics data, here focusing on a Visium platform [ 22 ] dataset from the middle temporal gyrus–which is a vulnerable region in the early stage AD [ 66 ]. For each cortical layer, we computed differential gene expression between AD and control samples and further estimated TF deregulation using pyPAGE.

As expected, 13 of the TFs regulons that we previously identified as associated with AD in the bulk data showed significant deregulation in at least one cortical layer in the spatial analysis with 7 of them showing significant deregulation among all layers ( Fig 4E ). Intriguingly, many of these TFs exhibit distinct spatial patterns of activation of their corresponding regulons. For example, SOX10, DUX4, and CDX2 regulons are preferentially activated in deep cortical layers, while FOXP2 and NR2F1 regulons are active in the middle layers. While most of these associated TF regulons were activated in AD compared to control samples, the KDM5A regulon exhibits strong downregulation across all layers in AD samples. Overall, this analysis shows that transcriptional programs affected by AD have different levels of regional specificity.

Systematic identification of RNA-binding proteins associated with pathological post-transcriptional gene regulation in AD

In the previous sections, we explored AD-associated transcriptional programs. Following, we performed a similar analysis of post-transcriptional programs controlled by RBPs and miRNAs. To do that, we computed unbiased estimates of transcript stability from MSBB RNA-seq data using our previously developed tool REMBRANDTS [ 27 ] and then assessed differential stability between AD patients and healthy donors. Noteworthy, in the analysis of post-transcriptional regulation it is critical to use estimates of RNA turnover, which is controlled by RNA-binding proteins and microRNAs, and cannot be substituted by differential gene expression, since it mainly reflects effects of transcriptional regulators. Further, we compiled and collated an annotation of post-transcriptional regulons using multiple sources: namely ENCODE [ 67 ] and POSTAR databases [ 68 ], as well as the Mireya Plass study [ 69 ]. By applying pyPAGE to this new comprehensive annotation of RBP targets we were able to disentangle the contribution of each RBP to the changes in stability and uncover novel post-transcriptional regulators associated with AD.

Utilizing the framework described above, we identified 16 candidate RBPs that are associated with differential RNA stability in AD ( Fig 5A ). Several of these proteins have been previously implicated in AD pathology. For example, in an earlier study, we had identified ZFP36 as a regulator of AD-related changes in RNA stability in an independent dataset [ 27 ]. Here, pyPAGE confirmed that ZFP36 activity is indeed perturbed in AD as evidenced by upregulation of the genes targeted by this RBP. Similarly, pyPAGE identified deregulation of HNRNPK and HNRNPD, two RBPs that are silenced in AD due to epigenetic alterations [ 70 , 71 ]. The HNRNPK regulon is destabilized in AD samples, indicating that this RBP acts as an enhancer of RNA stability. In contrast, HNRNPD promotes the degradation of its regulon, which results in an increase in their stability in AD [ 70 , 72 ]. The ability of pyPAGE to capture known and novel RBP associations with AD not only confirms the utility of this approach, but also provides readily testable hypotheses around the underlying modes of regulation.

(A) Heatmap representation of RBP regulons that are differentially expressed between AD and non-AD which we identified using pyPAGE. Here rows correspond to RBPs and columns to gene bins of equal size ordered by differential stability, the cells are colored according to the enrichment of genes from regulons in a corresponding bin. The leftmost column of this heatmap represents the differential expression of RBPs themselves. (B) Various roles performed by the identified RBPs based on the analysis of scientific literature. In this representation colored cells represent a recorded association between a protein and corresponding mechanism of action. (C) Deregulation patterns of the miRNA target gene-sets identified by pyPAGE. *miR-506 targets with GTGCCTT in their 3’ untranslated region. (D) Differential activity of RBP and miRNA regulons in various brain cell types. The codes for the analyzed cell-types: neurons (Neur), astrocytes (Ast), oligodendrocytes (Oli), oligodendrocyte progenitor cells (Opc), microglia (Mic). Differential activity of RBP regulons was estimated based on differential rates of RNA splicing and degradation. miRNA regulons were analyzed using only estimates of degradation rates. In these heatmaps significant associations (p-value<0.05) are marked by colored frames. Regulation pattern is estimated as normalized conditional mutual information of the association multiplied by the sign of log fold change.

https://doi.org/10.1371/journal.pcbi.1012346.g005

To show relevance of the inferred associations in the context of AD, we explored biological functions of the identified RBPs. First, we looked at the mechanisms they use to control RNA metabolism ( Fig 5B ) . Based on existing scientific literature, most of our candidate RBPs seem to regulate RNA stability directly upon binding either through repelling or recruiting RNA degradation factors [ 73 – 81 ]. However, some of these RBPs employ other mechanisms, such as HNRNPK which stabilizes RNA by preventing inclusion of cryptic exons during alternative splicing–a process that is critical in several neurodegenerative diseases [ 82 ]. Further, we performed analysis of overrepresented biological pathways within regulons of the identified RBPs to characterize their role in biological processes ( S8A Fig and S1 Table ) . This analysis revealed that these RBPs play a crucial role in regulating stability of various transcriptional regulators and genes involved in molecular signaling, including those associated with G protein-coupled receptor activity and kinase interactions. Differential activity of these RBPs in AD might in part explain observed deregulation of transcriptional programs via direct or indirect interactions. Additionally, the analysis of pathway enrichment within RBP regulons showed that a handful of the identified RBPs such PABC4, DROSHA, RPS3 and LARP4 are involved in the regulation of oxidative pathways which might be the root cause of AD-related mitochondrial dysfunction [ 83 ] and metabolic deficiencies [ 84 ].

Identification of post-transcriptional programs governed by microRNAs in Alzheimer’s disease

MicroRNAs are another major class of post-transcriptional regulators that have been previously associated with AD pathology [ 85 ]. Though many miRNAs have been reported to change their expression in the context of AD [ 85 ], there has been no systematic effort to infer their activity from changes in expression of their downstream targets. We sought to fill in this gap by applying pyPAGE to analyze AD-related coordinated changes in RNA stability of all annotated sets of miRNA targets, and identified eight candidate miRNA regulons ( Fig 5C ). In most cases AD-specific deregulation patterns of miRNA target genes were characterized by an increase in their stability. This suggests reduced activity of the miRNAs, which is in line with the notion that AD neurons lose the ability to process miRNAs [ 86 ]. Six out of eight identified miRNAs were previously associated with AD and its pathological mechanisms [ 87 – 91 ]. Functional analysis of genes regulated by these miRNAs ( S8B Fig and S1 Table ) showed that they largely control neuronal processes (neurogenesis, differentiation and signaling). Taken together, the marked deregulation of the identified miRNAs combined with their potential roles in AD and/or brain function makes them an attractive list of targets for further focused research.

Cell-type specific analysis of post-transcriptional regulatory programs in AD

Similar to our characterization of cell type specific patterns of transcriptional regulons, we next assessed the activity of RBP- and miRNA-driven regulons in each of the disease-relevant cell types. For this, we analyzed single cell RNA-seq data from Morabito et al. [ 21 ], which contains cells from both neuronal and glial lineages. Since RNA stability is difficult to estimate directly from single cell RNA-seq data, we used functions from the ‘velovi’ package [ 92 ] for inference of RNA splicing and degradation rates within each cell type and compared healthy and AD donors. These estimates present an unbiased view of post-transcriptional regulation by controlling for the effect of transcriptional activity. Given their known regulatory functions, we used both differential stability and splicing rates to test for differential RBP activity. For miRNAs, on the other hand, we focused primarily on differential stability rates.

This analysis confirmed that the post-transcriptional regulatory programs we had previously identified in bulk RNA-seq data were predominantly deregulated in cells of glial lineage, and to a lesser degree in neurons ( Fig 5D ). Oligodendrocytes, astrocytes, and oligodendrocyte progenitors predominantly showed an increase in splicing rates and a decrease in degradation rates of RBP regulons in AD samples compared to controls. Neurons, on the other hand, did not exhibit broad changes in splicing rates and only a few RBP regulons showed changes in their degradation rates. Results from miRNA regulons, however, were less concordant between the single-cell and bulk data modalities. Four out of eight miRNAs we identified above showed significant deregulation patterns, and only in two cell types. We observed a significant decrease in the stability of miR-106b-5p and miR-20A-5p regulons in microglia. Interestingly, upregulation of miR-106b-5p marks microglial activation [ 93 ], suggesting that its potential role as a regulator of inflammatory response in AD. Astrocytes showed a significant increase in the stability of miR-124-3p and miR-506-3p regulons, which are known to be critical for astrocyte transdifferentiation into neurons [ 94 ] and neuronal proliferation [ 95 ], respectively. In line with the importance of glia in AD pathology, overall our data highlights that most prominent patterns of miRNA and RBP regulon perturbations occur in non-neuronal cells.

pyPAGE identifies deregulation of post-transcriptional programs as a major source of heterogeneity within the cohort of Alzheimer’s patients

Finally, we employed pyPAGE to describe clinical implications of deregulation of the previously discovered master regulators. For that, we performed an analysis of their activity in each individual sample, measured using conditional mutual information (CMI) values that are provided by pyPAGE as part of its output. Here, the information criterion was estimated between the target genes and the standardized gene expression/transcript stability profile of each sample. As mentioned earlier, one of the benefits of using CMI is that it both captures monotonous changes in regulon abundance ( S9A Fig ) as well as non-linear patterns, which allows pyPAGE to efficiently measure regulon activity in individual samples.

Based on estimated activities of transcriptional and post-transcriptional regulons, we used Cox regression to assess associations between regulon activities and patient survival. Interestingly, we found that post-transcriptional programs are most strongly associated with patient survival, represented mainly by RBP regulons (LIN28A, HNRNPD, EWSR1, FXR1, HNRNPK, ILF3, PCBP1) and a few miRNA regulons (miR-302C-5P and miR-6825-5P) ( S9C Fig ). Using activity estimates of these RBP regulons, we performed unsupervised clustering of samples. This resulted in two clusters ( Fig 6A ), showing starkly different levels of activity of several RBPs. Remarkably, in this stratification low activity of the selected RBP regulons was associated with shorter survival (p-value = 0.007 of the Kaplan–Meier (log rank) test ( Fig 6B ) and hazard ratio of 4.84 (p-value = 0.012) of the Multivariate Cox model ( Fig 6D )). In comparison, no other major biological covariate, such as age, sex and cognitive state, had a significant effect on survival ( Fig 6D ). Next, we established that the activity estimates of RBPs in samples from a group with shorter survival was similar to those observed in healthy samples ( Fig 6C ). This observation indicates that modulation of these post-transcriptional programs is likely a compensatory mechanism that helps mitigate effects of AD progression. Following this analysis we asked whether the samples from the group with low RBP activity can be further stratified based on the activity of the target RBPs. In a new stratification of these samples we found two sub-clusters, among which the one with higher activity showed increased survival as compared to the subcuster with lower RBP activity (p-value = 0.037 of the Kaplan–Meier test) ( S9D Fig ). These results highlight the underexplored significance of post-transcriptional regulation in AD and their potential clinical utility in improving patient stratification.

(A) Heatmap representing differences in the activity of the previously identified post-transcriptional regulons in AD samples. Factors which activity is significantly associated with survival are underscored. The dendrogram reflects the results of unsupervised clustering of samples based on activity of factors associated with survival. (B) Kaplan-Meier curve representing the difference in survival between two groups of patients stratified based on the activity of post-transcriptional regulons. (C) Comparison of the activity of selected RBP regulons in healthy samples and samples from two AD clusters. (D) Summary of the Cox regression analysis.

https://doi.org/10.1371/journal.pcbi.1012346.g006

In order to identify underlying genetic root causes of the observed heterogeneity among AD patients, we performed differential analysis of genomic variants enrichment between the two identified groups of patients ( Table A in S1 Text ). Specifically, we looked at the AD-related GWAS [ 96 ] polymorphisms and polymorphisms related to genes of the analyzed RBPs. In this manner we have identified that patients with a longer lifespan are characterized by AD-related genomic variants in the upstream regions of CD55, PLK2 and SAP18 protein-coding genes. On the other hand, patients in the second group are characterized by variants upstream of STC1 and SCARA3 genes. The analysis of mutations associated with the RBPs showed that longer lifespan is associated with an upstream variant of PABPC4 gene and conversely shorter lifespan is related to two intronic variants in FXR1 and LARP4 genes. Incidentally, among these RBPs only FXR1 shows differential activity patterns between samples which makes it a primary subject for further research of the proteins which influence AD survival time.

Understanding the patterns of gene expression regulation in a biological system is a challenging task, but one which holds the potential to provide valuable insights into key factors controlling the behavior of a biological system. In this paper, we propose a systematic approach for exploration of gene regulation based on the analysis of regulon activity, which we implemented in our framework called pyPAGE. This new framework addresses two major problems that complicate our understanding of gene expression: technical variation and the combinatorial action of multiple regulatory factors. First, by focusing on the differential activity of regulons rather than individual genes, pyPAGE becomes more robust to random variations in data. Second, by conditioning the association of gene expression with a particular regulon on the number of regulons each gene is associated with, pyPAGE effectively uses the structure of an interaction network to more efficiently disentangle perturbation responses of multiple factors. In settings with simulated data, pyPAGE outperforms other commonly used tools and demonstrates its robustness to noise in the input, such as data sparsity. We show that pyPAGE can be used for inference of changes in regulatory programs both on transcriptional and post-transcriptional levels and in bulk and single-cell data modalities.

In this study, we have showcased applications of pyPAGE to large RNA-seq datasets, both single-cell and bulk, in the context of Alzheimer’s disease. We were able to recapitulate a number of previously known regulons and uncover new insights into the potential role of multiple TFs, RBPs and miRNAs in AD. Deregulation of the target genes controlled by these regulatory programs might explain pathological changes in AD, as they control key biological processes such as cell cycle progression, cellular signaling, neurogenesis, proteostasis and mitochondrial functions. By using single cell data of healthy and AD samples, we were able to capture the cell type specific patterns of the identified regulon that otherwise remained obscure in the analysis of bulk data. We found that AD-associated deregulations in gene expression in neurons are modulated primarily at the transcriptional level, whereas glial gene expression reprogramming is predominantly due to aberrant activity of post-transcriptional regulators. We further leveraged pyPAGE for the inference of regulatory programs hidden in the bulk data and uncovered characteristic upregulation of developmental TF regulons in most brain cell types. Similarly, we have identified that miR-124-3p, one of the key miRNAs that is depleted in AD brains, most strongly affects astrocytes, potentially blocking their transdifferentiation into neurons [ 94 ].

Like most complex human diseases, AD is a heterogeneous disease, likely with multiple molecular subtypes. Our results indicate that the activity of the various regulatory programs identified can be used to risk-stratify AD patients. In our stratification, patients with higher activity of several RBPs, in particular those associated with oxidative pathways, showed overall better clinical outcomes. As new therapeutic strategies against AD emerge, knowledge of these regulatory programs may allow for better stratification of responders and non-responders among the patient population.

Materials and methods

Analysis of bias in gene-set annotations.

To demonstrate specific properties of gene-set membership distribution, we analyzed several commonly used annotations of gene-sets. Specifically, we examined annotations of biological pathways and miRNA targets [ 38 ] from MSigDB [ 3 , 97 ], TF regulons from Vorontsov et al. [ 37 ] and RBP regulons which we compiled ourselves. For each gene in an annotation we, first, calculated the number of gene-sets each gene is a member of, then, divided these values into several equally sized bins and then showed that this distribution can be closely approximated with the power law. In each case we report the inferred gamma parameter of the distribution and the R 2 statistics. We observed that for several annotations the long tail of the gene-set membership distribution deviates from the power law, indicating bias in annotations.

We further explored the possible sources of bias observed in annotations. Firstly, we analyzed the association between the number of protein interactions and the protein abundance. To determine the number of interactions, we utilized the STRING database [ 98 ] considering only physical interactions with a confidence score greater than 0.9. As a source of protein abundance data we used proteome measurements from HEK-293 cells [ 99 ]. Secondly, we analyzed the association between pathway membership and citation index [ 100 ] for each gene in the annotation of biological pathways. In both cases we were able to show that there is a linear relationship between the analyzed variables indicated by high correlation values.

pyPAGE algorithm

The computational framework implemented by pyPAGE includes three key steps. First, it preprocesses the input data by discretizing the differential gene expression profile into equally sized bins and converting a gene-set annotation into a binary format. Second, it performs estimation of conditional mutual information between differential gene expression and each gene-set. Here the metric is conditioned on the gene-set membership profile which is also discretized into several bins.

In our algorithm, X stands for a set of bins in the expression vector (by default 10 bins), Y for a set {0, 1} which represents presence or absence of a gene in a regulon and Z is a set of bins (by default its size is 3) into which we discretize gene-set membership. Prior to computation of the metric, no probability smoothing is performed. In order to speed up our method, we implemented an efficient computation of the metric using numba package [ 101 ].

Following estimation of conditional mutual information, the program estimates the significance of the inferred associations. For this, a non-parametric testing procedure is implemented. The differential expression profile is randomly shuffled multiple times and at each iteration the conditional mutual information is calculated between these profiles and the scrutinized gene-set. Then the significance of the association is inferred as the proportion of cases when the CMI of randomly permuted differential expression profile turned out to be greater than CMI of the original profile. This procedure is applied to a list of gene-set in a descending order of the CMI calculated at the second step. When multiple insignificant associations (by default 5) come in a row, the testing stops.

The final output of pyPAGE consists of gene-sets that, first, exhibit significant association with expression represented by high CMI and, second, pass a redundancy test. The resulting list can be additionally filtered using multiple testing correction methods. However, in practice we observe that all gene-sets that pass previous tests pass the Benjamini–Hochberg procedure. In the end results are reported in the form of a table. It also has an option of the graphical output, which is a heatmap representing gene-set members enrichment within equally sized bins ordered by differential gene expression. To quantify the level of over- and under-representation in each bin, we employ the hypergeometric test to calculate two distinct p-values: one for the likelihood of observing fewer genes from the same gene-set in a given bin (under-representation), and another for the probability of observing more genes (over-representation). If the p-value for over-representation is lower than that for under-representation, we consider the gene-set to be over-represented in that bin; otherwise, it is considered under-represented. The minimal p-value is then log-transformed, and multiplied by -1 if the gene-set is over-represented.

pyPAGE is implemented as a python package and is openly available at https://github.com/goodarzilab/pypage .

In order to validate various aspects of pyPAGE execution we have developed a series of tests based on simulated data. Our main focus was on proving that the program is robust to gene-set annotation biases. To do that we evaluated its average performance using artificial datasets and benchmarked it against other programs. We also conducted two additional tests to demonstrate consistency of pyPAGE predictions. Though to produce highest performance pyPAGE needs to receive clean data with strongly manifesting biological signals, we still expected the program to perform decently in various adversarial cases. By subsampling the original input and by modulating the robustness of signals encoded in the data, we wanted to show at which cost the reduction in input quality is to the overall performance. The full description of the tests is provided further.

Simulating differential expression

All implemented tests required simulating differential gene expression data. Though in principle pyPAGE does not depend on any particular type of the input distribution, we decided to make the input data realistic. So, we based the simulation on the actual gene expression estimates from a real biological system, which in this case was activation of murine CD4 T cells (PRJNA296380) [ 102 ]. Starting off with fastq files, we first applied Salmon [ 103 ] to quantify gene counts and afterwards applied DESeq2 [ 68 ] to estimate differential gene expression between two groups of samples. To produce an arbitrary amount of simulated gene expression profiles using this data, we performed random shuffling of the data across genes. This procedure minimized the informational content which had been in the data initially. We demonstrated that by performing pyPAGE gene-set enrichment analysis of terms from biological pathway annotation using original and augmented data. While the analysis of the original data reported 76 pathways, using simulated data pyPAGE on average reported only 3.36 pathways (standard deviation = 1.80). Though this procedure removes almost completely any informational content, it preserves the properties of the original data distribution. Moreover it can be repeated multiple times to produce many samples of simulated differential expression profiles.

Simulating pathway annotations

In addition to the differential expression profiles, each test needed a simulated pathway annotation. To be able to assess the performance of pyPAGE at later stages, during annotation design whether a gene-set is deregulated or not should be directly encoded in its structure. And so our goal was to create annotations each containing in total 1100 gene-sets, among which 100 are known to be deregulated in the corresponding differential gene-expression profile. The number of genes that went into a newly designed gene-set chosen based on the distribution of sizes of TF regulons [ 37 ]. To simulate unperturbed gene-sets we randomly picked the necessary number of genes. To imitate deregulated gene-sets, we used genes that are either over- or under-expressed in the corresponding simulated data; the details of simulating such gene-sets are described in the following paragraph.

When simulating deregulation our objective was to imitate patterns observed in real systems, so as to avoid making the task of predicting gene-set deregulation too easy or too difficult. In a real case scenario genes of a deregulated gene-set are not always exclusively found among the most overexpressed or underexpressed genes, rather the signal is distributed over the whole range of values. Likewise, in our simulation we built probabilistic models to control which genes go into a newly formed gene-set. We aimed that the resulting distribution of differential gene expression in each gene-set should match the truncated normal distribution on an interval between -1 and 1 with predefined parameters. So for each gene-set we first defined a distribution of expected differential gene expression, then randomly sampled a number of values from it and selected those genes from a simulated differential expression profile which most closely matched these values. Parameters of the expected distribution were also selected randomly. For an upregulated gene-set, its mean was picked randomly from a random distribution in an interval between 0.5 and 1, for a downregulated gene-set the interval was between -1 and -0.5. The standard deviation was uniformly selected in the range between 0.1 and 0.2. To simulate dual regulation, we mixed genes selected for an upregulated gene-set with those of a deregulated one. Finally, to imitate biological and technical noise to each gene-set we added a randomly selected group of genes. The ratio of randomly added genes was controlled by signal-to-noise ratio, which in the general case was set to 33%.

The resulting gene-set annotations were unbiased with gene-set membership following approximately normal distribution. To produce biased annotations, we needed to add bias externally. And so we selected 2000 genes which would represent genes with promiscuous activity. Then each gene-set in an annotation was concatenated with additional 100 genes randomly selected from this group. The idea behind this approach is that added genes would occur much more frequently than all other genes effectively creating bias in gene-set membership distribution.

Benchmarking pyPAGE

Using the artificial datasets containing simulated differential expression and gene-set annotations, we were able to analyze pyPAGE effectiveness and compare it with other programs: iPAGE and fgsea. The test was basically a classification task: each program needed to predict deregulated gene-sets. The performance was judged based on the knowledge of true patterns of regulation which were encoded in simulated annotation. We used multiple metrics to assess performance of programs which were averaged across 20 iterations with different sets of simulated data. To present these results we plotted ROC (receiver operating characteristic) curves, PR (precision recall) curves, DET (detection error tradeoff) curves, FDR (false discovery rate curves) curves and calculated respective AUC (area under the curve) scores. Since the number of unperturbed gene-sets was much greater the most relevant metric in this task was PR-AUC score. Programs were tested in 4 different conditions to reflect the effect of bias and non-linear regulation on the quality of predictions.

Robustness test

Since in practical applications, the input data of a gene-set can often be corrupted, we developed two tests to explore how robust pyPAGE is to the augmentation of the input. The first strategy that we implemented for simulating input perturbation was increasing the amount of noise in the data. In the previous we set signal-to-noise ratio to 33%, however here we varied this parameter in the range between 0 and 100%. The second type of augmentation we used was implemented by randomly subsampling the input data. We iteratively decreased the number of genes in the input by 5% measuring pyPAGE performance. Both tests output pyPAGE sensitivity changes in response to input modulation.

Analysis of Alzheimer’s disease

Finally, we used these estimates to perform pyPAGE analysis of TF regulons deregulation using gene-sets from Vorontsov et al. [ 37 ] (full description of the annotation is provided in Table B in S1 Text ). For this run we used the following set of parameter values: alpha 0.01, redundancy ratio 0.2, number of expression bins is 10 and the number of bins into which the gene-set membership vector is split is 3. The overall summary of this and following pyPAGE experiments can be found in Table C in S1 Text .

In order to analyze post-transcriptional regulation, we first needed to estimate differential transcripts stability. To do this we first applied Salmon [ 103 ] to quantify separately exonic and intronic counts and then applied REMBRANDTS [ 27 ] to produce unbiased estimates of stability for each sample. Next using t-test we compared stability between pathological and normal samples; the resulting p-values and fold changes were used to quantify differential stability similarly to how we estimated differential gene expression. Finally, based on these estimates pyPAGE was able to infer deregulation patterns of RBP regulons from our newly compiled annotation and microRNA targets [ 38 ] from MSigDB [ 3 , 97 ] ( Table C in S1 Text ). pyPAGE analysis of RBPs was done with alpha set to 0.05, redundancy ratio 0.3, number of expression bins 10, gene-set membership bins 3. miRNAs were analyzed using number of expression bins 10, gene-set membership bins 3, alpha 0.01 and no redundancy filtering, instead we selected only those miRNAs whose targets were significantly enriched either among top differentially upregulated genes or top differentially downregulated genes. This was done because we wanted to limit our analysis to the strongest signals in the data.

Cell-type specific deregulation patterns were estimated using single cell data. For this purpose we used two datasets. In both cases we relied on prior annotation of cell-types provided by the authors of the studies. One was an already processed dataset presented by the ROSMAP study [ 20 ]. The dataset contained 70634 cells from 48 donors and was balanced in respect to pathology and sex covariates. We used it for the analysis of TF regulons. There within each cellular type cluster we applied t-test to estimate differential gene expression between cells from AD and non-AD samples. Then pyPAGE was used to estimate deregulation of TF regulons within each cell type. To analyze activity of the regulators that we previously detected in our analysis we set redundancy ratio to 0.0 and alpha to 1.0, this prevented pyPAGE from truncating its input only to significant associations. In addition we also ran pyPAGE with redundancy ratio being set to 0.3 and alpha to 0.05 which allowed us to identify TFs whose perturbation remained hidden in the bulk data. For the analysis of RBP and miRNA deregulation, we used another single cell RNA-seq dataset from Morabito et al. [ 21 ]. In this data there were 18 samples in total of which 11 were from AD patients (6 female, 5 male) and 7 from healthy donors (2 female, 5 male). In total there were 61472 cells. We processed this dataset de novo using cellranger [ 105 ] and velocyto [ 106 ] in order to finally run velovi [ 92 ]. We applied velovi to single cell datasets of individual samples in order to estimate patient specific rates of RNA splicing and degradation in each cellular type. In the end in a manner similar to how we estimated differential gene expression, here we used t-test to estimate differential kinetic rates between patients with and without AD. Based on the differential splicing and degradation rates, using pyPAGE we were able to infer the activity of RBP regulons. To analyze miRNA regulons, we used only differential degradation rates. With RBPs and miRNAs we only wanted to get estimates of differential activity of the factors identified in bulk analysis, so in all of the cases we set redundancy ratio to 0.0 and alpha to 1.0.

Spatial analysis of TF deregulation patterns was performed using sequencing data from the Visium platform of the middle temporal gyrus [ 22 ]. In total the dataset had 25293 (22085 after filtering out ones with less than 2000 gene counts) spots from 6 cortical slices (3 healthy and 3 AD). The count data was scaled and then log normalized after which we computed differential gene expression between AD and control samples using the Wilcoxon test. This was done for each cortical layer relying on their prior annotation. Further, we applied pyPAGE to estimate differential activity of TFs that we previously identified as associated with AD. In the end for each TF in each cortical layer we report its deregulation pattern computed as normalized CMI multiplied by the sign of log fold change and its significance.

We also analyzed the association between the activity of the identified regulons and patients’ survival. To do this, we analyzed another bulk RNA-seq dataset from the ROSMAP study [ 107 ] stored in the Synapse database under syn21589959. The samples were from dorsolateral prefrontal cortex and posterior cingulate cortex. We considered only samples from AD patients for whom the age the disease was first diagnosed was not censored. This left us with 46 samples. For these samples we performed gene expression quantification using salmon and then normalized these estimates. We then used this normalized gene expression to quantify activity of TF regulons in each sample using pyPAGE estimates of conditional mutual information. Similarly, we used salmon and REMBRANDTS to estimate transcripts stability in each sample and afterwards used normalized stability to infer activity of RBP and miRNA regulons. Activity of regulons was estimated on a per-sample basis, so that instead of differential expression/stability, pyPAGE was run with normalized expression/stability values for each sample. Next, we used regulon activity in Cox Proportional-Hazards Model to identify factors associated with survival of AD patients. Based on this we selected regulons which were significantly associated with survival. We then analyzed patterns of activity of these regulons in various samples and effectively excluded those regulons which did not show any marked difference in their activity between different samples. Then based on the activity of the remaining regulons we performed clustering of samples which resulted in two big clusters. Next, we used this stratification to analyze its association with survival. For this, we built Kaplan-Meier curve and performed Cox Proportional-Hazards Model which incorporated other AD relevant covariates such as sex, APOE genotype, Braak score, Cerad score.

For the two clusters that we identified in our survival analysis we looked at the polymorphisms that characterize these groups using a complementary WGS dataset (syn2580853). In particular we considered 2 categories: those that are associated with AD polymorphisms from GWAS catalog [ 96 ] and those that are associated with the RBPs from our analysis. We report variants that are enriched in one group of samples (present in over 80% of samples) or depleted in another (present in fewer than 20% of samples).

Characterization of regulon functions

We describe the functions of the identified TF, RBP and miRNA regulons based on the enrichment of molecular pathways among their members. To do that, we used a statistical overrepresentation test from PANTHER [ 108 ], which includes multiple testing corrections. This test estimated the significance of pathway enrichment within a set of all genes that are targets of a specific factor. Then we aggregated the inferred functions of each regulon and presented them in a heatmap form.

For RBPs we also characterize their roles in various RNA metabolic processes. This was done based on manual curation of scientific literature. We describe RBP associations with following molecular mechanisms: alternative splicing [ 109 – 116 ], RNA editing [ 117 – 119 ], polyadenylation [ 120 ], RNA export [ 121 , 122 ], stress granule formation [ 123 , 124 , 124 – 129 ], RNA-decay [ 72 , 130 – 132 ], RNA stabilization [ 73 – 81 ] and miRNA functions [ 132 – 137 ].

ENCODE processing

The ENCODE eCLIP data [ 67 ] was available for two cell lines: HepG2 and K562. In total, there were 234 <RBP, cell type> pairs (687 samples including replicates, and matched controls for each RBP). RBPs with no matched control or with matched control only (without the experiments) were excluded from the analysis resulting in the final set of 222 <RBP, cell type> pairs. The data were analyzed as follows: (1) reads were preprocessed as in the original eCLIP pipeline [ 138 ], (2) trimmed reads were mapped to the hg38 genome assembly using hisat2 [ 139 ], (3) the aligned reads were deduplicated [ 138 ], (4) the uniquely and correctly paired reads were filtered with samtools [ 140 ], (5) gene-level read counts in exons were obtained with plastid [ 141 ] using gencode v31 comprehensive gene annotation [ 142 ], therefore joint read counts were obtained for genes with overlapping annotations, (6) differential expression analysis against matched controls was performed with edgeR [ 26 ], (7) Z-Scores were estimated for each gene across data for all RBPs for the fold change values estimated in (6). Based on (6) and (7) reliable RNA targets of each RBP were defined as those passing 5% FDR, raw log2FC > 0.5, Z-Score(log2FC) > 1.645. The processed data is stored in S2 and S3 Tables.

Supporting information

S1 fig. bias in gene-set annotations..

(A) Characteristics of a gene-set membership degree distribution in the TF regulons annotation. The top plot represents the observed distribution with a power law function being fit to it. We also report the gamma parameter of this distribution and the R 2 The bottom plot represents the deviation of the observed distribution from the power law. (B) Similar representation for RBP regulons. (C) Similar representation for miRNA targets. (D) Similar representation for GO pathways.

https://doi.org/10.1371/journal.pcbi.1012346.s001

S2 Fig. Performance of pyPAGE benchmarked against iPAGE and fgsea in various simulated conditions.

For comparison we used multiple metrics, results are presented as ROC, PR, DET curves and plot of dependency between alpha and FDR.

https://doi.org/10.1371/journal.pcbi.1012346.s002

S3 Fig. Variation in expression of TF regulons.

(A) Boxplots representing expression of TF regulons which were identified using pyPAGE in AD and non-AD samples from female and male donors. (B) Boxplots representing expression of the same TF regulons in AD and non-AD samples in different age cohorts.

https://doi.org/10.1371/journal.pcbi.1012346.s003

S4 Fig. Top 100 of pathways enriched in TF regulons.

(A) Heatmap representing biological pathways that are enriched among targets of TFs we have identified in our analysis. Here we present top 50 pathways which are significantly enriched in most of the regulons excluding several generic terms, like “molecular process”. (B) This heatmap represents another 50 pathways significantly associated with the analyzed TFs.

https://doi.org/10.1371/journal.pcbi.1012346.s004

S5 Fig. Cell-type specific patterns of TF target genes deregulation in AD.

(A) Differential activity of TFs in excitatory neurons inferred using pyPAGE based on the analysis of single-cell RNA-seq data. (B) TF regulons deregulated in inhibitory neurons. (C) TF regulons deregulated in astrocytes. (D) TF regulons deregulated in oligodendrocyte progenitors. (E) TF regulons deregulated in microglia. (F) TF regulons deregulated in oligodendrocytes. (G) TF regulons deregulated in endotheliocytes. (H) TF regulons deregulated in pericytes.

https://doi.org/10.1371/journal.pcbi.1012346.s005

S6 Fig. Correlation between the expression of RBPs and their regulons’ average stability.

The barplot representing the correlations between the expression of RBPs and the average stability of their regulons. Asterix indicated significant correlation (p-value<0.05).

https://doi.org/10.1371/journal.pcbi.1012346.s006

S7 Fig. Variation in expression of TF regulons.

(A) Boxplots representing stability of RBP regulons which were identified using pyPAGE in AD and non-AD samples from female and male donors. (B) Boxplots representing stability of the same RBP regulons in AD and non-AD samples in different age cohorts. (C) Boxplots representing stability of miRNA targets within AD and non-AD samples from female and male donors. (D) Boxplots representing stability of miRNA targets within AD and non-AD samples from different age cohorts.

https://doi.org/10.1371/journal.pcbi.1012346.s007

S8 Fig. Top 100 pathways enriched in RBP and miRNA regulons.

(A) Heatmap representing biological pathways that are enriched among targets of TFs we have identified in our analysis. Here we present only the top 100 pathways which are significantly enriched in most of the regulons. (B) Similar representation for miRNA targets.

https://doi.org/10.1371/journal.pcbi.1012346.s008

S9 Fig. Association of expression of a subset of HOX genes with survival of patients with AD.

(A) Comparison of the similarity between various regulon activity metrics, namely conditional mutual information (CMI), mean regulon abundance and the expression of a factor itself. (B) Heatmap representation of activity of transcriptional regulons in different patients. (C) Bar Plot representing significance of association of activity of each previously identified regulon with patient’s survival. (D) Kaplan-Meier curve representing the difference in survival between two groups of patients stratified based on the activity of post-transcriptional regulons within the group with low RBP regulon activity.

https://doi.org/10.1371/journal.pcbi.1012346.s009

Table A in S1 Text . Differential enrichment of genomic variants in clusters identified based on RBP activity. Table B in S1 Text . Description of gene-set annotations. Table C in S1 Text . Summary of the pyPAGE experiments.

https://doi.org/10.1371/journal.pcbi.1012346.s010

S1 Table. Functions of the identified regulators.

https://doi.org/10.1371/journal.pcbi.1012346.s011

S2 Table. Processed ENCODE K562 eCLIP data.

https://doi.org/10.1371/journal.pcbi.1012346.s012

S3 Table. Processed ENCODE HepG2 eCLIP data.

https://doi.org/10.1371/journal.pcbi.1012346.s013

Acknowledgments

We thank Andrey Buyan and Ivan V. Kulakovskiy for providing reprocessed ENCODE eCLIP data. We also thank Dr. April Pawluk for reviewing and editing this paper.

The results published here are in whole or in part based on data obtained from the AD Knowledge Portal ( https://adknowledgeportal.org/ ). These data were generated from postmortem brain tissue collected through the Mount Sinai VA Medical Center Brain Bank and were provided by Dr. Eric Schadt from Mount Sinai School of Medicine.

• View Article
• PubMed/NCBI
• Google Scholar
• 23. Entropy, Relative Entropy, and Mutual Information. In: Elements of Information Theory [Internet]. John Wiley & Sons, Ltd; 2005 [cited 2022 Jun 3]. p. 13–55. Available from: https://onlinelibrary.wiley.com/doi/abs/10.1002/047174882X.ch2
• 101. Lam SK, Pitrou A, Seibert S. Numba: a LLVM-based Python JIT compiler. In: Proceedings of the Second Workshop on the LLVM Compiler Infrastructure in HPC [Internet]. New York, NY, USA: Association for Computing Machinery; 2015 [cited 2024 Mar 30]. p. 1–6. (LLVM ‘15). Available from: https://dl.acm.org/doi/10.1145/2833157.2833162