Oncology databases for predicting tumor radiosensitivity and chemosensitivity: a critical translational synthesis

Abstract

Background . Oncology databases have become a foundation for developing models of tumor sensitivity to drug and radiation therapy; however, their clinical value is determined not by data volume but by the quality of the response phenotype, external validation, and evidence of treatment-effect modification. Objective. To perform a critical translational synthesis of oncology databases and computational resources used to predict tumor radiosensitivity and chemosensitivity. Materials and methods . A narrative critical review was conducted across five groups of resources: clinico-molecular atlases, pharmacogenomic panels, radiosensitivity databases, functional dependency maps, and clinical and infrastructure platforms. The response phenotype, reproducibility, external validation, transferability across model systems, risk of bias, and clinical interpretability were assessed. Results . The greatest translational maturity is currently demonstrated by homologous recombination deficiency in the context of DNA-damaging therapy and genomic-adjusted radiation dose as a research predictor of radiotherapy benefit. Contemporary DepMap resources, clonogenic radiosensitivity databases, organoid models, radiomics, and federated learning expand validation opportunities but do not eliminate the gap between association and clinical utility. Conclusions . Models should undergo cross-database, external clinical and, where feasible, prospective validation. Control of batch effects, cohort representativeness, and phenotype standardization are particularly important. The immediate practical role of oncology databases is to prioritize biomarkers and support study design rather than autonomously replace clinical decision-making.