High-throughput genetic screenings to improve cancer variant interpretation and uncover context-dependent effects

Cancer is driven by genetic alterations, yet the functional consequences of most mutations remain poorly understood. In this thesis, I first outline the genetic foundations of cancer and the challenges associated with interpreting variants. I then apply computational approaches to analyze large-scale genomics datasets, investigating the effects of genetic alterations with the aim of refining variant interpretation and identifying novel biomarkers to better guide precision oncology. Using genome-wide CRISPR and RNAi loss-of-function screenings, I developed ButterflyVI, a framework to systematically annotate over 20,000 cancer-associated variants into four functional categories: oncogene dependency, tumor suppressor dependency, mutation tolerance, and bypass-of-essentiality. This approach revealed unexpected patterns, including vulnerabilities linked to heterozygous loss-of-function variants and context-specific bypass-of-essentiality, where loss-of-function mutations in essential genes seem to confer a selective advantage only in particular cellular contexts. Importantly, ButterflyVI also enabled the discovery of novel predictive biomarkers such as RPL5 loss for MDM2 inhibition, which was validated experimentally. In addition to this central work, I contributed to two independent collaborative projects. The first involved a pan-cancer analysis of lineage-factor dependence, identifying subsets of cells resistant to lineage-factor inhibition and revealing alternative transcriptional programs shaping cancer vulnerabilities. The second integrated co-mutation patterns with machine learning to uncover combinations of mutations predictive of metastatic progression, offering insights into tumor evolutionary dynamics and potential prognostic biomarkers. Together, these studies demonstrate that genetic alterations act in a highly context-dependent manner. By combining functional genomics, computational modeling, and large-scale datasets, this work enhances variant interpretation and identifies actionable vulnerabilities, providing critical insights to guide precision oncology.