Drug Repositioning for Schizophrenia and Depression/Anxiety Disorders: A Machine Learning Approach Leveraging Expression Data

Development of new medications is a lengthy and costly process, and drug repositioning might help to shorten the development cycle. We present a machine learning (ML) workflow to drug discovery or repositioning by predicting indication for a particular disease based on drug expression profiles, with a focus on applications in psychiatry. Drugs that are not originally indicated for the disease but with high predicted probabilities serve as candidates for repurposing. This approach is widely applicable to any chemicals or drugs with expression profiles measured, even if drug targets are unknown. It is also highly flexible as virtually any supervised learning algorithms can be used. We employed the ML approach to identify repositioning opportunities for schizophrenia as well as depression and anxiety disorders. We applied various state-of-the-art ML approaches, including deep neural networks (DNNs), support vector machines (SVMs), elastic net regression, random forest, and gradient boosted trees. The predictive performance of the five approaches in cross validation did not differ substantially, with SVM slightly outperforming the others. However, other methods also reveal literature-supported repositioning candidates of different mechanisms of actions. As a further validation, we showed that the repositioning hits are enriched for psychiatric medications considered in clinical trials. We also examined the correlation between predicted probabilities of treatment potential and the number of related research articles, and found significant correlations for all methods, especially DNN. Finally, we propose that ML may provide a new avenue to exploring drug mechanisms via examining the variable importance of gene features.