SEPIA: Polypharmacy side-effect prediction for combinations of multiple drugs
Finn Lueth, David Wagemann, Cheng-Wei Liao, Leon Rauschning, Christian Romberg, Julius Schmidt, Jannis Arnold · 2023
Polypharmacy refers to the simultaneous administration of multiple (>= 5) medications. While polypharmacy is a necessary part of treating multimorbid or vulnerable patients, it is associated with the risk of adverse drug interactions, non-adherence, and a decline in quality of life. The risk profile of polypharmacy is non-linear and poorly understood. In care settings, polypharmacy can affect more than half of all patients; in total, about 15% of the US population is considered to be polypharmaceutical – a rising trend in the context of an aging population. Clinical or laboratory testing of every possible combination of drugs is infeasible. Computational prediction of polypharmacy side effects offers a much more scalable approach to reducing the risk associated with polypharmacy. However, previous approaches have only predicted pairwise drug interactions. Here, we present SEPIA, a multimodal network-based multidrug interaction prediction tool. SEPIA computes embeddings for each drug from a multimodal network containing known drug-drug interactions, drug-protein and protein-protein interactions, and their chemical structures. SEPIA then combines the embeddings using a set-transformer architecture to give a side effect prediction for a set of simultaneously administered drugs. Using SEPIA, we hope to reduce the burden of polypharmacy and foster improved clinical decision-making.