Unraveling determinants of drug response: PRISM viability profiling of hundreds of oncology therapeutics reveals mediators of selectivity and mechanism

Incomplete understanding of drug mechanism, selectivity, and polypharmacology can contribute to failed clinical trials. Drug candidates are generally characterized with limited pre-clinical tools and a narrow indication focus based on expected target biology. Using approaches such as large-scale PRISM profiling of 900 cell lines that have been genomically profiled in the Dependency Map coupled with machine learning analysis can provide a more thorough and agnostic understanding of oncology drugs. Here, we demonstrate that systematic characterization of an Oncology Reference library of over 250 oncology drugs, encompassing biologics (e.g., cytokines, antibody-drug conjugates), targeted protein degraders, and small molecules, of which two-thirds have never been tested with large-scale cell line profiling, enables better understanding of on-target and off-target effects. We developed models to quantify selectivity, on-target activity, and polypharmacology. Surprisingly, in some instances, these measures of cellular specificity can anticipate differential outcomes in clinical trials. Among unexpected findings, we identified a recurrent viability signature predictive of compound lysosomal accumulation, a phenomenon associated with phospholipidosis. In addition, CDK4-selective inhibition emerged as a potential therapeutic vulnerability across a diverse range of cancer types, including CDK6-deficient or CDK4-altered cancer. Our findings establish the Oncology Reference dataset and PRISM as a benchmark for evaluating current and emerging cancer therapies: combining large-scale systematic cell line profiling, genomic characterization, and machine learning analysis enables comprehensive characterization of oncology drugs with potential to inform clinical development.