NetLapMDA: Network-Based Laplacian Regularized Least Squares Method for miRNA-Disease Association Prediction

Numerous investigations have highlighted a robust relationship between microRNAs (miRNAs) and various illnesses. To tackle the limitations of traditional experimental techniques, which are both time-intensive and resource-heavy, and to boost validation accuracy, significant attention has been directed toward developing computational frameworks for predicting miRNA-disease associations. In this study, we introduce a novel method, termed Network-based Laplacian regularized least squares for miRNA-disease association prediction (NetLapMDA). This approach separately computes the similarities between miRNAs and diseases before combining them into an integrated framework. The resulting model is systematically optimized and assessed through leave-one-out cross-validation, five-fold cross-validation, and case study of three specific diseases. Performance comparisons with existing methodologies reveal that the proposed model achieves an AUC of 0.9213 in global LOOCV, 0.8606 in local LOOCV, and 0.9181±0.00049 in five-fold cross-validation. Additionally, the top 50 miRNAs identified in the case studies were fully validated by experimental evidence. NetLapMDA adopts a semi-supervised strategy that utilizes the shared characteristics of miRNAs and diseases by employing a normalized Laplacian operator. It defines and solves a predictive objective function to estimate association scores and demonstrates remarkable accuracy in forecasting miRNA-disease relationships. The model effectively enhances predictive power for previously unexplored associations by leveraging the smooth continuity of classifiers within similarity metrics.