Integrating lncRNA data for prediction of miRNA-disease association using network fusion and matrix completion.

MicroRNAs (miRNAs) regulate essential biological processes and play critical roles in the pathogenesis of complex human diseases. Consequently, accurate identification of potential miRNA-disease associations (MDAs) is of great importance for disease diagnosis, prognosis, and therapeutic development. However, traditional experimental approaches are often time-consuming and costly. Although numerous computational methods have been proposed to address these challenges, many of them suffer from severe data sparsity and an inability to predict associations for novel entities, such as miRNAs or diseases with no prior known links. Moreover, most existing models overlook the important mediating role of long non-coding RNAs (lncRNAs) in disease-related regulatory mechanisms. To address these challenges, we propose a novel computational framework based on network fusion and matrix completion for miRNA-disease association prediction. The proposed approach integrates heterogeneous biological information, including miRNA-disease, lncRNA-disease, and miRNA-lncRNA associations, together with disease semantic similarity and miRNA/lncRNA functional similarity. Specifically, a three-layer heterogeneous network is constructed, and an unbalanced random walk strategy is employed to propagate information across network layers, effectively alleviating the sparsity of the original association matrix. Subsequently, a matrix completion strategy is applied to infer potential associations and generate final prediction scores. Comprehensive experiments using 5-fold cross-validation and leave-one-out cross-validation demonstrate that the proposed method achieves AUC values of 0.9745 and 0.9935, respectively, outperforming several state-of-the-art approaches. Furthermore, case studies on major human diseases confirm the robustness, reliability, and practical applicability of the proposed framework.