ButterflyVI: enabling high-throughput variant interpretation and biomarker discovery with functional genomics

Abstract

Characterizing the functional and therapeutic relevance of cancer mutations is a primary goal and a major challenge in precision oncology. Whereas predictive approaches exist, they often lack functional validation, which cannot scale to large numbers of cancer-associated variants. Here, we integrated statistical learning with experimental evidence from high-throughput functional screenings to annotate >20000 unique variants. We defined a variant as functional if it altered the effect of gene loss, and classified four possible outcomes: oncogene and tumor suppressor dependencies, mutation tolerance, and bypass-of-essentiality. Up-to-60% of variants annotated as functional were previously considered of unknown significance. Paradoxically, bypass-of-essentiality was common among loss-of-function (LoF) variants at several tumor suppressors, including VHL, ARID1A, and RBM10. In these cases, loss of the wild-type genes was deleterious, independently of the tissue of origin, but not when they already harbored recurrent LoF variants, suggesting loss of these tumor suppressors provides a context-specific advantage. Using our annotations, we discovered somatic variants that increased sensitivity to loss of therapeutically targetable genes, representing new candidate biomarkers. Among these, we validated RPL5 LoF mutations as a biomarker of response to selective MDM2 inhibitors. A dedicated web portal (butterflyvi.unil.ch) enables exploration of all variant annotations and candidate biomarkers. This study highlights the potential and need to expand large-scale functional screenings to empower variant interpretation in the clinic.