Common and Differential Enrichment across Multiple Experiment Profiles: A Multivariate Bayesian Variable Selection Approach

High-throughput gene profiling measures the expression of thousand of genes simultaneously. Gene set enrichment analysis examines the pre-defined, biologically meaningful sets of genes with increased power and robustness to find subtle changes. With gene expression data measured by multiple profiles, such as in different cell lines and under different drug treatments, it is of great interest to elucidate which and how the enrichment differs among experiment profiles. We conceptualize the common and differential enrichment with cross-profile mean and variance for the coefficients in the regression model. A Bayesian hierarchical model is developed to select variables corresponding to common and differential enrichment. To evaluate the performance of our model, we applied it to simulated data with various level of common and different enrichment, and compare the variable selection accuracy to the model with no distinction of enrichment type. Finally, we demonstrate that our model is generally applicable to parametric gene set enrichment analysis by analyzing KEGG pathways in two cancer studies.