Enhancing Drug Repositioning Through Collaborative Metric Learning: A Novel Approach

Inthe era of big data, leveraging computational approaches for drug repurposing emerges as a promising and efficient method to uncover novel applications for existing medications. This methodology bears significant potential for advancing precision medicine. Deviating from traditional drug development methods, drug repositioning emerges as a costeffective and low-risk strategy. Framed as a top-K recommendation task, the drug repositioning challenge involves aligning medications with the most probable diseases using data related to drugs and diseases. In this work, collaborative metric learning is applied to predict the top diseases that can be recommended to drugs. Firstly, pre-processing of the drugdisease association dataset obtained from CTD database is done, in addition to obtaining drug related data from the same. Secondly, sampling of the dataset is done using SMOTE technique to address the imbalance issue caused due to far lesser number of drug-disease pairs with known associations than the pairs with no known associations in the dataset. Finally, using collaborative metric learning in conjunction with the balanced drug and disease related data, novel drug-disease interaction (DDI) associations are predicted. Latent vectors of drugs and diseases are obtained precisely, and the association probability of drug-disease pairs is then determined using them. According to the experimental analysis, the method works better than the most advanced medication repositioning strategies.