Cross-modal imputation and gated GCN for predicting miRNA-disease association (CIGGNET)

microRNA(miRNA) is a short-chain non-coding RNA molecule encoded by endogenous genes. Currently, many miRNAs related to complex diseases have been found, which provides help for further exploring the molecular mechanism of disease pathogenesis. We proposed an algorithm named CIGGNET for predicting the association between miRNA-disease based on cross-modal data imputation and gated graph convolution network. First, CIGGNET uses a cross-modal data imputation operation on the miRNA-disease association matrix to obtain the filled association matrix. Second, CIGGNET integrates miRNA-disease heterogeneous networks, extracts features of miRNAs and diseases use random wander algorithm, and learns miRNA and disease embeddings using graph convolutional network. Third, CIGGNET uses a gating operation to select the appropriate convolution layer. The control gate adaptively outputs suitable convolution layers based on the similarity of different convolution layers and scores unobserved associations. The mean AUC of CIGGNET is 0.9423 in 100 five-fold cross-validations.