Explainable drug side effect prediction in central neural system via biologically informed graph neural network

The early detection of potential side effects (SEs) is a critical yet formidable challenge within the realms of drug development and patient healthcare management. Conventional in-vitro or in-vivo approaches for SE detection are often not feasible to scale during the preclinical phase for numerous drug candidates. Innovations in explainable artificial intelligence offer the prospect of early detection of potential SEs for novel therapeutics prior to their release in the market, as well as the explication of the underlying biological mechanisms. In this context, we present a novel biologically informed graph-based model, called HHAN-DSI, which capitalizes on multimodal interactions among molecular entities. Applied within the domain of the central nervous system (CNS) - the organ system associated with the largest number of SEs - our model demonstrates its capability to reveal previously unrecognized SEs of various psychiatric drugs. Moreover, HHAN-DSI elucidates the associated biological mechanisms, delineating an intricate network of genes, biological functions, drugs, and SEs.