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

The study introduced a model called WNMFDDA, which used weighted non-negative matrix factorization to predict potential associations between drugs and diseases. It aimed to improve drug repositioning by addressing limitations like overfitting and sparse associations. The model incorporated graph Laplacian and Tikhonov regularization for accuracy and was tested on a dataset with 1,933 associations between 593 drugs and 313 diseases. It achieved high AUC values of 0.939 and 0.952 in ten-fold cross-validation, outperforming other methods. Despite challenges like sparse data, WNMFDDA effectively identified true associations and showed promise as a tool for biomedical research.