Improving a molecular signatures database for use with gene set enrichment analysis : using co-expression to split and redefine functional gene sets

Thesis (M. Eng.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2007.Includes bibliographical references (p. 89-90).Manually curated gene sets related to a biological function often contain genes that are not tightly co-regulated transcriptionally. which obscures the evidence of coordinated differential expression of these gene sets in relevant experiments. To address this problem, we explored strategies to refine the manually curated subcollection of the Molecular Signatures Database (MSigDB) for use with Gene Set Enrichment Analysis (GSEA). We examined the manually curated gene sets in context of an atlas of gene expression of many normal human tissues. To refine gene sets, we clustered the genes in each set based on co-expression across the tissues to produce more tightly co-regulated children gene sets that are also likely more accurate representations of the biological process or processes described by the gene set. We evaluated the performance of the clustering algorithms by refining gene sets in the context of several published GSEA analyses and verifying that the children gene sets score higher with GSEA than do the parents. We created and annotated a new, refined version of a large portion of the manually curated component of MSigDB, which we hope will be a resource for the GSEA community.by Yuliya Kodysh.M.Eng