Modeling the relationship between gene expression and mutational signature.

Abstract

Mutational signatures computed from somatic mutations, allow an in-depth understanding of tumorigenesis and may illuminate early prevention strategies. Many studies have shown the regulation effects between somatic mutation and gene expression dysregulation. We hypothesized that there are potential associations between mutational signature and gene expression. We capitalized upon RNA-seq data to model 49 established mutational signatures in 33 cancer types. Both accuracy and area under the curve were used as performance measures in five-fold cross-validation. A total of 475 models using unconstrained genes, and 112 models using protein-coding genes were selected for future inference purposes. An independent gene expression dataset on lung cancer smoking status was used for validation which achieved over 80% for both accuracy and area under the curve. These results demonstrate that the associations between gene expression and somatic mutations can translate into the associations between gene expression and mutational signatures.