Abstract A central goal of cancer research is the identification of cancer genes that drive tumour growth and progression. Existing approaches to this problem typically leverage frequentist approaches based on patterns of somatic mutagenesis in DNA. Here, we interrogate RNA variant allele frequencies to identify putative cancer genes with a novel computational tool, RVdriver , from bulk genomic-transcriptomic data within 7,948 paired exomes and transcriptomes across 30 cancer types. An elevated RNA VAF reflects a signal from multiple biological features: clonal mutations; mutations retained or gained during somatic copy-number alterations; mutations favoured by allele-specific expression; and mutations in genes expressed preferentially by the tumour compartment of admixed bulk samples. RVdriver , a statistical approach that classifies RNA VAFs of nonsynonymous mutations relative to a synonymous mutation background, leverages this information to identify known, as well as putatively novel, cancer genes, with comparable performance to DNA-based approaches. Furthermore, we demonstrate RNA VAFs of individual mutations are able to distinguish ‘driver’ from ‘passenger’ mutations within established cancer genes. Low-RNA VAF EGFR mutations otherwise annotated as drivers of glioblastoma by DNA tools harbour a phenotype of reduced EGFR signalling, whilst high-RNA VAF KDM6A mutations otherwise annotated as passengers exhibit a driver-like H3K27me3 expression profile, demonstrating the value of our approach in phenotyping tumours. Overall, our study showcases a novel approach for cancer gene discovery, and highlights the potential value of multi-omic and systems-biology approaches in finding novel therapeutic vulnerabilities in cancer to bring about patient benefit.

RNA allelic frequencies of somatic mutations encode substantial functional information in cancers