DLRNMFMDA: miRNA-Disease Associations Prediction Method Based on Dual Laplacian Regularized Non-Negative Matrix Factorization

microRNAs (miRNAs) represent a class of non-coding RNAs intricately linked to numerous biological processes and human diseases. Traditional biological methods for identifying miRNA-disease associations are both labor-intensive and time-consuming. Consequently, computational approaches for uncovering these associations have emerged as a prominent area of research. Addressing the limitation that most existing models fail to effectively utilize information from unvalidated samples, this study introduces a novel prediction framework that incorporates Dual-Laplacian Regularization into a Non-negative Matrix factorization method for MiRNA-Disease Associations prediction (termed DLRNMFMDA), leveraging diverse biological data. The proposed semi-supervised model integrates several innovative features, and enhances prediction accuracy by incorporating negative sample information, employs the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$L_{2}$</tex> -norm to mitigate noise interference for more robust predictions, and applies Laplacian regularization within a non-negative matrix factorization framework to capture intrinsic data structures more precisely. These enhancements significantly bolster model performance and offer new analytical approaches for complex data. An area under the ROC curve of 0.9112 was achieved through five-fold cross-validation, with all top 50 miRNA molecules from validation results corroborated by experimental evidence. These findings underscore the reliability of the DLRNMFMDA model.