A machine-learning framework has been developed to identify robust drug biomarkers using pharmacogenomic data from three-dimensional organoid culture models. This approach successfully identified biomarkers that accurately predict drug responses in colorectal cancer patients treated with 5-fluorouracil and bladder cancer patients treated with cisplatin. The biomarkers were further validated using external transcriptomic datasets of drug-sensitive and -resistant isogenic cancer cell lines. Additionally, concordance analysis between the transcriptomic biomarkers and independent somatic mutation-based biomarkers confirmed the method’s validity. This work offers a promising method for predicting cancer patient drug responses by leveraging pharmacogenomic data from organoid models and employing gene modules and network-based approaches.