Identify Disease-Associated MiRNA-miRNA Pairs Through Deep Tensor Factorization and Semi-supervised Learning

MicroRNAs (miRNAs) are a class of small non-coding RNAs that play a significant regulatory role in the development of disease. Researchers have explored a variety of computational methods to predict the association between miRNA and disease, which accelerates the discovery of biomarkers. However, current studies mainly focus on binary relations, ignoring the impact of synergistic miRNAs on disease. Moreover, the acquisition of negative sets also hinders the improvement of relevant algorithms. To address these issues, we propose a novel tensor-based framework called TF-SSL to predict disease-associated miRNA-miRNA pairs. Reliable negative samples are extracted by combining semi-supervised learning and deep clustering. Afterward, TF-SSL utilizes self-supervised graph attention aggregation layer to efficiently represent node features over complicated biological networks. The learned features of miRNA and disease are used to reconstruct the association tensor for discovering possible triple relationships. Empirical results showed that the proposed method achieved state-of-the-art performance under five-fold cross-validation. Case studies on three complex diseases further demonstrated the effectiveness of TF-SSL in identifying potential disease-related miRNA-miRNA pairs.