Fig. 1: Pan-cancer RNA correlates of patient survival.
a Kaplan–Meier plot of TCGA patient overall survival (OS) by cancer type12, indicating the need for cancer type correction in pan-cancer analyses. b Using different statistical cutoffs, the numbers of RNA transcripts significantly associated with patient survival are indicated (by Cox), with or without correction for cancer type. c Based on their tumor transcriptome, TCGA patients are stratified according to a pan-cancer RNA signature of worse survival (FDR < 1%, Cox correcting for cancer type), with a Kaplan–Meier plot showing the differences in survival outcome. d A top set of 9555 RNA transcripts (out of 60,660) were significantly associated with patient survival (FDR < 5%, Cox correcting for cancer type) in pan-cancer analyses across 10,271 cancers. For each individual cancer type, the respective associations with patient survival are indicated for the 9555 RNAs. Also indicated are the numbers of RNAs significantly associated with survival for each cancer type in analyses restricted to the given cancer type. e Pan-cancer survival associations (by Cox correcting for cancer type) were determined for 433 proteins measured by Reverse-Phase Protein Array (RPPA)29. The overlaps of the RPPA-based survival correlates with the RNA survival correlates corresponding to each gene are represented, considering worse versus better survival correlates separately. Genes listed individually are well-established cancer-associated genes by COSMIC57. Enrichment p values by one-sided Fisher’s exact test. f The RNA pan-cancer survival signature from part (d) was applied to eight independent tumor gene expression datasets outside of TCGA (bladder26, breast31, liver32, lung adenocarcinoma33, ovarian34, pediatric brain tumors25, prostate35, renal36) and in each instance was able to stratify patients according to survival outcome.
