Drug–disease associations prediction via Multiple Kernel-based Dual Graph Regularized Least Squares

Abstract Predicting associations in drug–disease network provides effective information for the drug repositioning. Therefore, it is an important task to develop an effective drug–disease association prediction method. In this paper, we propose a model, called the Multiple Kernel-based Dual Graph Regularized Least Squares Model (MKDGRLS), to predict potential drug–disease associations. First, we calculate the multiple kernels of drug and disease spaces respectively. Moreover, we utilize multiple kernels and related Laplacian regularization terms to construct MKDGRLS. Finally, we solve the object function of MKDGRLS by the Alternating Least squares algorithm (ALSA), for predicting drug–disease associations. Simultaneously, we design a variant MKDGRLS-A, which is added one dimension as the bias for compensate the inexact solution of linear systems. Our proposed approach has better prediction performance than existing prediction tools under two types of cross validation on the three real drug–disease association network datasets. The results of the case studies of the two diseases prove that our model can effectively predict new associations. We also test MKDGRLS on six real-world networks and it achieves better performance than other methods. In conclusion, our research work can effectively discover potential drug–disease associations, and can provide effective help for related drug repositioning work.