Fast graph convolutional models incorporating matrix factorization for predicting microbe-disease associations

Abstract

Abstract Revealing the relationship between microbe and disease is of great significance to the diagnosis, treatment, and prevention of disease. To overcome the expensive cost and trial-and-error settings, a series of in-silico methods have been proposed to predict microbe-disease association. However, the predictive performance of the current methods is modest. In this paper, we propose a new computational method based on Fast Graph Convolutional and Matrix Factorization, called FGCNMF, which addresses microbe-disease association prediction as a binary classification task by learning embedding representation of nodes on a microbe-disease network. We integrate background information from both microbe and disease spaces into the same global network framework, and use the randomized Singular Value Decomposition algorithm to obtain high-quality initial embedding representations of node. Then, Fast Spatial Convolution is implemented to enhance the embedding representations. Finally, using the enhanced representation of node pairs as input, and using Extra-Trees classifier to predict the final label. Experimental results demonstrate that FGCNMF has improved performance in comparison with other state-of-the-art computational methods on the benchmark datasets.