Relating Individualized Disease Network Centrality to Cancer Gene Essentiality
Genome-wide CRISPR knockout screens have demonstrated variable gene essentiality across diverse cancer cell lines. However, understanding the molecular biology underlying these variations remains a challenge. Network biology has enabled crucial insights into biological systems such as the centrality-lethality hypothesis, postulating that hubs in gene networks tend towards essential genes. We hypothesize that gene centrality in cell-line specific disease gene networks can help explain variable cancer gene essentiality. Here we propose S-PERMUTOR, a supervised adaptation of the PERsonalized Mutation evaluaTOR framework (Weiskittel et al. 2021) for deriving individualized networks using known phenotypic labels. We then construct novel networks in lung, breast, and kidney cancer cell lines and build node-centrality trained machine learning models to predict cancer gene essentiality across a panel of 100 genes. We demonstrate that disease network centrality, not necessarily of the gene under study, can effectively predict gene essentiality in individual cancer cell lines. Combining model derived features with individual network topologies, we dissect potential mechanisms underlying gene essentiality and demonstrate the unique biological insights enabled by S-PERMUTOR.