Gene-to-Protein-to-Disease: a segmentation-based approach to identify new cancer prognostic signatures of exome sequence variants

Abstract

Somatic mutations are early drivers of tumorigenesis and tumor progression. However, the mutations typically occur at variable positions across different individuals, resulting in the data being too sparse to test meaningful associations between variants and phenotypes. To overcome this challenge, we devised a novel approach called Gene-to-Protein-to-Disease (GPD) which accumulates variants into new sequence units as the degree of genetic assault on structural or functional units of each protein. The variant frequencies in the sequence units were highly reproducible between two large cancer cohorts. Survival analysis identified 247 sequence units in which somatic mutations had deleterious effects on overall survival, including consensus driver mutations obtained from multiple calling algorithms. By contrast, around 75% of the survival predictive units had been undetected by conventional gene-level analysis. We demonstrate the ability of these signatures to separate patient groups according to overall survival, therefore providing novel prognostic tools for various cancers. GPD also identified sequence units with somatic mutations whose impact on survival was modified by the occupancy of germline variants in the surrounding regions. The findings indicate that a patients genetic predisposition interacts with the effect of somatic mutations on survival outcome in some cancers.