Prediction of miRNA-disease Association Based on Multi-Source Inductive Matrix Completion

MicroRNAs (miRNAs) are endogenous non-coding RNAs of about 23 nucleotides in length that play important roles in a variety of cellular biochemical processes. A large number of studies have demonstrated that miRNAs are involved in the regulation of many human diseases. Accurate and efficient prediction and identification of the association between miRNAs and human diseases will have great significance for the early diagnosis, treatment and prognosis assessment of human diseases. In this paper, we propose a model called Autoencoder Inductive Matrix Completion (AEIMC) to identify potential miRNA-disease associations. Specifically, we first capture the interaction features of miRNA-disease associations based on multi-source similarity networks, including miRNA functional similarity network features, miRNA sequence similarity features, disease semantic similarity features, disease ontology similarity features, and Gauss interaction spectral kernel similarity features between disease and miRNA. Secondly, autoencoders are used to capture more complex and abstract data representations of miRNA and disease. Finally, the learned high-level features are used as inputs to the induction matrix completion model to obtain the miRNA-disease association prediction matrix. At the end of the artical, an ablation experiment was performed to confirm the validity and necessity of introducing miRNA sequence similarity and disease ontology similarity for the first time.