Overlapping Group Logistic Regression with Applications to Genetic Pathway Selection.

Area of Science:

• Genomics and Bioinformatics
• Statistical Genetics
• Computational Biology

Background:

• Identifying key genes for specific phenotypes is a major challenge in genome-wide expression analysis.
• Pathway-based analyses like gene set enrichment analysis (GSEA) are common for hypothesis testing but not regression.
• Overlapping pathways and lack of software hinder the integration of pathway information into regression models.

Purpose of the Study:

• To develop a novel regression approach for analyzing gene expression data with overlapping pathway structures.
• To extend the existing R package grpreg to accommodate overlapping group structures using a latent variable approach.
• To compare the performance of the new method against ordinary lasso and GSEA.

Main Methods:

• Developed grpregOverlap, an extension of the grpreg package, to handle overlapping group structures via a latent variable model.
• Utilized simulated and real gene expression datasets for comparative analysis.
• Compared the proposed method with ordinary lasso and gene set enrichment analysis (GSEA).

Main Results:

• The grpregOverlap method successfully incorporates overlapping pathway information into regression models.
• Incorporating prior pathway information significantly enhances the accuracy of gene expression classifiers.
• Demonstrated key differences between hypothesis-testing (GSEA) and regression approaches for pathway data analysis.

Conclusions:

• The developed grpregOverlap package provides a robust solution for regression analysis of gene expression data with overlapping pathways.
• Integrating pathway information via this regression framework substantially improves the predictive accuracy of gene expression classifiers.
• The study highlights the distinct advantages and applications of regression versus hypothesis-testing approaches in pathway-based genomic analysis.