Wasserstein-Based Similarity Constrained Matrix Factorization for Drug-Drug Interaction Prediction

We introduce a modification to Manifold Regularized Matrix Factorization for predicting drug-drug interactions, employing a Wasserstein similarity measure. Our findings demonstrate that this measure offers enhanced accuracy compared to previously tested similarity metrics. Notably, it exhibits resilience to variations in the latent feature vector size, denoted as k, enabling the projection of the drug-drug interaction matrix onto a significantly reduced dimensional space, thereby reducing computational costs. The proposed method holds significant implications for clinical practice by highlighting drug-drug combinations warranting further investigation for potential interactions.