Cold-start problems in data-driven polypharmacy side effect prediction

Combining drugs, a phenomenon called polypharmacy, can induce (new) adverse side effects. The identification of adverse combinations is a key task in pharmacovigilance. In this context, in silico approaches based on machine learning are promising as they can learn from a limited number of combinations to predict for all. In this work, we identify various polypharmacy prediction tasks and show how they can improve post-market surveillance systems or detect polypharmacy side effects earlier during drug development. We elaborate on model validation and propose a new model that obtains AUC-ROC=0.843 for the hardest “cold-start” task up to AUC-ROC=0.957 for the easiest task.