A Statistical Method For Identifying Chemical-Genetic Interactions Using Linear Mixed Models

Abstract

This research is focused on building statistical solutions to identify chemical-genomic interactions. We have used a linear mixed model to study the trend in the abundances of the genes in a population when exposed to varying concentrations of drugs. In this model, each influence of the drug on each individual gene is treated as random effects. For every gene, our model yields a gene-specific slope for the abundance vs the concentration of the drugs. These slopes are then subjected to 1-sided test to determine the genes with the most significantly outlying slopes, i.e., giving us an insight on the potential target of the drug under consideration. To gain further insights, we have used the GSEA analysis on the ranks of the slopes to understand the impact of the drugs on a pathway of genes. The developed model is validated on the publicly available chemical-genomics dataset published by the Broad Institute and multiple hypomorph libraries created by our collaborators.