Two patients, the same diagnosis, the same drug — and opposite outcomes. Today’s digest is about teaching machines to predict, from a tumor’s molecular profile, whether a drug will actually work. Iorio et al.’s GDSC laid the foundation, mapping alterations from ‘11,289 tumors from 29 tissues’ onto ‘1,001 molecularly annotated human cancer cell lines … correlated with sensitivity to 265 drugs.’ Chang et al.’s CDRscan predicted ‘drug effectiveness from cancer genomic signature.’ Chiu et al.’s DeepDR was honest about the hard part: ’the translation into predicting drug response in tumors remains challenging.’ Sharifi-Noghabi et al.’s MOLI fused multiple omics, since ‘integrating additional omics can improve the prediction accuracy.’ Liu et al.’s DeepCDR brought the drug in too, exploring ‘intrinsic chemical structures of drugs.’ And Kuenzi et al.’s DrugCell built ‘an interpretable deep learning model of human cancer cells’ — one that also predicts drug synergy. Precision oncology, learned.
