Trustworthy Drug-Drug Interaction Prediction Model via Weighted Aggregating Conformal Prediction

Drug-drug interaction (DDI) prediction is crucial for ensuring the safety and efficacy of combinationdrug therapies, flowever, most existing methods lack uncertainty estimation, limiting their reliabilityand practical adoption. While conformal prediction (CP) offers guaranteed coverage in classificationtasks, its direct application to DDI prediction faces challenges in achieving valid marginal coverageand elicient prediction set sizes. 'To overcome these limitations, this work introduces WeightedCPDDIthe first conformal prediction framework tailored for DDI tasks, WeightedCPDDI incorporates a nove.weighting mechanism to achieve user-specilic coverage and enhance prediction eliciency, Evaluationacross transductive and inductive settings demonstrates WeightedCPDDI's superior performance. Inthe transductive setting, it consistently achieves target coverage with significantly smaller predictionsets, showcasing improved calibration and eficiency, In the inductive setting, WeightedCPDDI robustlyoutperforms state-of-the-art methods in both seen-unseen and unseen-unseen scenarios. Its strongperformance across various backbone models highlights its generalizability and practical reliability forDDI prediction under uncertainty.