A Latent Mixture Model for Gene Set Analysis

• Activity Number: 105
• Type: Invited
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: IMS
• Author(s): Zhijin Wu*
• Companies: Brown University
• Keywords: genomics ; gene set ; gene expression ; latent mixture

Abstract

Biological processes are known to involve genes and gene products that function in harmony. Recent studies have demonstrated increased statistical power by combining information from genes in known pathways, as the signal on individual genes or SNPs may be weak but become stronger when aggregated. One common feature of most gene set analysis methods is that at the individual gene level, a location shift in expression is expected form of differential expression. We observe, however, in some biological systems, the treatment under study does not affect every relevant gene in an important gene set. Sometimes the probability that a relevant gene responds to the treatment can be low, such that the difference between the mean expression in the treated versus control is very small. Methods targeting the location shift as DE have very little power detecting these effects. We introduce a latent variable that is an indicator whether a gene responds to the treatment in a given sample, and target DE that are rare but with with large magnitude. We demonstrate great gain in statistical power in identifying gene sets that consist of these genes.