Predicting Drug-Disease Associations by Self-topological Generalized Matrix Factorization with Neighborhood Constraints

Predicting drug-disease associations (DDAs) is a significant part of drug discovery. With the continuous accumulation of biomedical data, multidimensional metrics about drugs and diseases are obtained, therefore how to effectively integrate them into computational models has become the focus of research. However, traditional methods only roughly integrate data without considering their differences. In this paper, we introduce a novel method for DDAs prediction based on self-topological generalized matrix factorization with neighborhood constraints (NSGMF). Instead of giving the same attention to each similarity metric, we perform data fusion with different information average entropy weights. And the fused data is used as constraint terms for matrix factorization to predict unknown DDAs. In addition, self-topological information is used to provide node feature indication in matrix factorization, which will effectively get rid of the problem that traditional matrix factorization is sensitive to external information. The experimental results of cross validation show that NSGMF method has better comprehensive performance than other DDAs prediction methods.