Fig. 1

Conceptual Overview of Research. a Somatic mutations, copy number alteration and gene expression data for each tumor were collected. b GeneRIF and gene summaries associated with genes were collected. c The semantic data associated with each gene was processed to create a word vector representation (note the differences in the word frequency profile for different genes). d A document representation for each tumor was created by combining the word vectors of each SGA associated with the tumor. e The document representations were used as input for a hierarchical topic model, which identified topics associated with each tumor. f The tumors were represented in topic space, and clustering analysis was applied to group tumors with similar topic allocations. g These clusters were then used to perform survival analysis on tumors of the same cancer type