Machine Learning for Identifying Overlap in Psychiatric and Neurological Drug Mechanisms

Psychiatric and neurological disorders often exhibit overlapping symptomatology and shared neurobiological mechanisms, yet pharmacological treatments are typically developed and administered in isolation. This research proposes a machine learning (ML) framework to identify potential “dual-use” drugs—compounds that may be therapeutically effective across both domains—by analysing drug-target interaction data. Drug-target profiles were curated from ChEMBL and Drug Bank, with each drug categorized as Psychiatric, Neurological, or Both. Feature vectors were constructed based on binary interactions with classified biological targets (e.g., GPCRs, ion channels, kinases). We applied unsupervised techniques such as t-SNE and UMAP for dimensionality reduction, which revealed distinct yet overlapping clusters of drug classes. Clustering analysis using the Elbow method and K-means (k = 5) further highlighted mechanistic intersections, particularly in the dual-use group. Supervised classification using a Random Forest model achieved an accuracy of 81%, though performance was affected by class imbalance. Feature importance analysis identified ion channels, kinases, and GPCRs as key predictors of dual-use classification, aligning with known cross-domain mechanisms in neuropharmacology. The visual and quantitative results collectively support the feasibility of using ML to uncover shared pharmacological pathways. This work lays a computational foundation for drug repurposing, reducing polypharmacy risks, and improving therapeutic strategies for comorbid psychiatric-neurological conditions. Future work will integrate clinical and genomic data to enhance model robustness and translational potential.