Improving reproducibility in pharmacogenomic screens through cross-study benchmarking

Drug-response measurements across pre-clinical pharmacogenomic studies remain poorly correlated, which limits biomarker discovery, precision oncology, and predictive modelling. The drivers of this inconsistency have been debated but not yet resolved. By integrating 15 pharmacogenomic studies encompassing 760 small-molecule compounds, 1,111 cell models, and 9.8 million dose-response measurements, we demonstrate that dose-response metric is the strongest driver of inconsistency, followed by experimental factors, such as treatment duration, plate format, and viability readout; in contrast, cell line molecular features contribute only minimally to reproducibility. Among drug classes, hormone therapies and PARP inhibitors show the highest concordance, whereas antimetabolites, topoisomerase inhibitors, and mitotic inhibitors exhibit substantial response variability across studies. To improve consistency, we developed a Drug Response Score (DRS), a proximity-weighted measure that emphasize pharmacologically informative concentrations near IC ₅₀ , and we demonstrate in systematic benchmarking how DRS markedly improved cross-dataset concordance. Applications to patient-derived neuroblastoma organoids and leukemia patients’ primary cells demonstrate that DRS improves replicate-level consistency in patients’ drug-response profiles. To improve reproducible pharmacogenomic studies, we make openly available an integrated Drug Response Resource (iDRR, https://aittokallio.group/iDRR/ ), a standardized 15-dataset portal that supports robust biomarker discovery and cross-study benchmarking.