Here we present a new MSigDB collection of “hallmark” gene sets and show how it can help to overcome these challenges. These hallmark gene sets are generated by a hybrid approach that combines an automated computational procedure with manual expert curation. The computational methodology identifies gene set overlaps and generates coherent representatives of them. The manual curation makes critical use of domain expert knowledge in order to: i) assign biological themes to groups of the original overlapping gene sets, ii) identify expression data for refinement and validation of the hallmark signatures, and iii) properly annotate the refined hallmarks. The hallmarks summarize information across multiple gene sets by emphasizing genes that display coordinate expression and represent well-defined biological processes, thereby reducing variation and redundancy, and providing a better delineated biological space for GSEA analysis.
