One-step Multi-view Inductive Matrix Completion for Gene-Disease Associations Prediction

Jie Xu, Lei Chen, Hui Wu, Siming Zha

Abstract: Discovering genes closely related to complex dis-eases is extremely important for disease diagnosis, treatment and prevention. Most existing methods use a two-step strategy with the beginning feature fusion step and the following inductive learning step to predict the potential gene-disease associations, ignoring the reciprocal relationship between the two steps which could help each other in achieving more accurate prediction results. In this paper, we propose a novel One-step Multi-view Inductive Matrix Completion (OMIMC) model, highlighted by joint latent representation learning and weighted PU (positive-unlabeled) learning, to predict gene-disease associations. Specifically, our model employ the latent representation learning scheme to simultaneously capture the consistency and complementary information of the sample with multi-view (even the sample with missing views), and thus obtain the common latent representations for genes/diseases. Furthermore, considering that the gene-disease associations prediction is essentially a PU learning problem, we also introduce the adaptive weighting scheme into traditional inductive matrix completion model to penalize the observed and the unknown associations differently. Finally, extensive experiments conducted on several real data set demonstrate the effectiveness of our proposed method.