Online Program

A Modified PageRank Algorithm for Pathway Prioritization Based on Projection Correlation and Generalized R-Squared (303707)

Keywords: Pathway ranking, PageRank algorithm, generalized R-squared, projection correlation, pathway dependencies

Identification of disease-related pathways is of great importance in revealing mechanisms of complex diseases as the pathway level analysis provides more functional insight than the single gene based analysis. Existing tools for pathway relevance ranking, however, are mostly based on enrichment analysis for each single pathway, which could be inaccurate by overlooking pathway dependencies. In this talk, we introduce a modified Google's PageRank framework to fully incorporate the pathway dependencies in pathway ranking. The new measurement well balances two aspects, namely the marginal correlation between pathways and a categorical phenotype and the correlations between pathways. We propose to use projection correlation to quantify pathway-pathway correlation, and a generalized R-squared for phenotype-pathway correlation based on a generalized linear model. Simulation study based on real biological pathways shows the superiority of the proposed method over marginal correlation and enrichment based methods.