A Weighted Non-Negative Matrix Factorization Approach to Predict Potential Associations Between Drug and Disease

The study introduces a model called weighted graph regularized collaborative non-negative matrix factorization (WNMFDDA) to predict potential associations between drugs and diseases. By calculating drug and disease similarities and using a weighted K nearest neighbor approach, the model reconstructs interaction score profiles. WNMFDDA demonstrated high predictive performance with AUC values of 0.939 and 0.952 on two datasets during ten-fold cross-validation, surpassing other methods. Case studies confirmed the model's predictions for several drugs and diseases, with a significant number of top candidate associations validated by existing databases. The results suggest that WNMFDDA is an effective tool for drug-disease association prediction, aiding biomedical research.