ORF screen analysis

The ORF screen (“toxicity”) analysis was performed in both a pan-cancer and tissue-specific manner. We repurposed the vehicle treatment arm of 17 independent ORFeome library screens that were conducted to assess drug sensitivity across eight tumor types (Genetic Perturbation Platform, Broad Institute). For each screen, cells were infected with the ORFeome pLX317 barcoded library (16,100 barcoded ORFs overexpressing 12,753 genes), selected with puromycin, and cultured for 14–21 days. We compared the high-throughput sequencing data from early (after puromycin selection) and late (after 14–21 days of cell culture) time points to estimate log₂-fold changes and significance in ORF barcode representation using a linear model, where each barcode measurement for a given gene in any screen was considered an independent observation. All screens were included for the pan-cancer analysis, whereas the tissue-specific analysis was limited to screens that share a cancer type. We estimated gene-set level differences using a linear model on the Wald statistic of each gene in a set.