Leveraging Uncertainty Estimates for Drug Response Prediction in Cancer Cell Lines

Machine learning models for drug response prediction in cancer cell lines carry the potential to advance precision oncology by tailoring treatments to the molecular tumor profile. Their application is challenged by variability in prediction quality and distribution shifts between training and application. Uncertainty estimation provides more information on the predictive distribution than point estimates, enabling comprehensive decision support and downstream analysis of predictions. Yet, the most effective uncertainty estimator is domain-specific. In this work, we benchmark uncertainty-aware models for drug response prediction. We focus on epistemic uncertainty via ensemble agreement, and aleatoric uncertainty via distributional modeling, or both. We find that ensemble-based estimates are more sensitive to distribution shift and can flag out-of-distribution examples. In contrast, distributional models yield stronger prediction error reductions among high-confidence subsets. Despite higher computational cost, the combination can provide both advantages: an ensemble of neural networks that estimate a Gaussian predictive distribution can reduce the mean squared error by 64 percent when restricting predictions to the 10 percent most confident drug-cell line pairs, and reliably indicates distribution shifts and platform differences. Beyond benchmarking, probabilistic predictions can identify drugs whose uncertainty bounds overlap with therapeutically relevant ranges. We also show that uncertainty estimates enable a new dimension of model interpretability: by attributing predicted uncertainty to input features, we identify genes that signal unpredictability of drug response rather than sensitivity or resistance. We further demonstrate uncertainty-guided selection of measurements for active learning. In summary, including uncertainty in drug response prediction supports better-informed model application. The code is available at https://github.com/PascalIversen/LUDRP.