Recommender System to Identify Drug-Target Associations Using Matrix Factorization
Sornsawan Phuangmalai, Apichat Suratanee, Kitiporn Plaimas · 2024
Recommender systems, widely used in information filtering systems, demonstrate efficacy in suggesting novel items to users. In this study, we propose the application of a recommender system to predict new drug-target associations. Here, drugs are represented as users, while protein targets are treated as items within the recommender system. However, due to sparse data in a matrix of drugs and protein targets, matrix factorization (MF) techniques are employed to decompose the extensive matrix into smaller matrices. We identify non-negative matrix factorization (NMF) and singular value decomposition (SVD) as top-performing models for this task. Subsequently, we integrated them into the traditional recommender system, encompassing both drug-based collaborative filtering and target-based collaborative filtering, to identify new drug-target associations aligned with observed interactions between drugs and target proteins. Finally, we evaluated the performance of our developed recommender system with matrix factorization for drug-target associations and compared the results with those obtained from a recommender system without matrix factorization.