Identification of Microbe-Drug Association based on Weighted Profile and Collaborative Matrix Factorization

Previous studies have shown that diseases are associated with microbe. To explore a more effective treatment for these diseases, unknown microbe-drug associations must be identified. However, existing models to identify microbe-drug association are limited. In our article, a predictive model (WPCMF) is presented for identifying microbe-drug associations based on weighted profile and collaborative matrix factorization. In WPCMF, the Gaussian Interaction Profile (GIP) can be used for computing the similarities of microbe and the drug, respectively. Then we use the Canonical SMILES of drugs to compute the chemical structures similarity of drugs. Two drug similarities are fused into an integrated drug similarity matrix. Weighted profile and collaborative matrix factorization are applied for predicting potential microbe-drug associations. Experimental results show that WPCMF achieves the average Area Under the Curve (AUC) values of 0.9096±0.0028, 0.9195±0.0019 and 0.9236 in 5-fold Cross-Validation (5 CV), 10-fold Cross-Validation (10 CV) and Leave-One-Out-Cross-Validation (LOOCV), respectively, which consistently outperforms other related methods (KATZHMDA, WP, CMF and Kron RLS). We think WPCMF is ideal as a supplement in the field of biomedical research.