Pathway crosstalk effects: Shrinkage and disentanglement using a Bayesian hierarchical model

Area of Science:

• Biomedical research
• Systems biology
• Bioinformatics

Background:

• Identifying biological pathways linked to clinical phenotypes is crucial.
• Classical over-representation analysis (ORA) methods suffer from issues like pathway overlap (crosstalk), leading to inaccurate pathway rankings and inflated error rates.
• The crosstalk phenomenon and its impact on ORA, including unreliable rankings for small pathways and increased type I and type II errors, have not been rigorously addressed.

Purpose of the Study:

• To develop a novel Bayesian hierarchical model to overcome the limitations of classical ORA.
• To provide more accurate pathway estimates and rankings by accounting for pathway overlap and other confounding factors.
• To reduce both type I and type II error rates in pathway analysis.

Main Methods:

• Development of a Bayesian hierarchical model.
• Incorporation of pathway overlap (crosstalk) into the analysis.
• Validation using both simulated and real biological datasets.

Main Results:

• The proposed Bayesian model yields more accurate pathway rankings compared to classical ORA.
• The method effectively addresses issues related to pathway overlap, small pathway unreliability, and multiple comparisons.
• Error rates (both type I and type II) are reduced, leading to more reliable results.

Conclusions:

• The Bayesian hierarchical model offers a more robust and accurate approach to pathway analysis in biomedical research.
• This method enhances the understanding of biological phenomena underlying clinical phenotypes.
• The developed R code and datasets are available for public use, facilitating further research.