A Meta-Learning Prediction Model for Identifying miRNA-Disease Associations Based on MAML

MicroRNAs (miRNAs) are a class of short non-coding RNA molecules, typically consisting of 20–22 nucleotides. These molecules regulate gene expression by binding to the mRNA of target genes, thereby reducing the expression of target gene products through complete complementary pairing and regulating the level and timing of gene expression through non-perfect complementary pairing. Furthermore, miRNAs play a role in the regulation of the occurrence and development of various diseases, including cardiovascular diseases. The abnormal expression of miRNA has been found to be closely associated with the development and progression of diseases. Consequently, the development of a reliable and accurate predictive model for identifying the association between miRNAs and diseases is of paramount importance. This study introduces a meta-learning model, MAMLMDA, that combines model-agnostic meta-learning (MAML) and CNN to identify the association between miRNA and disease. The model extracts and integrates the similarity between disease and miRNA based on experimentally verified association information between known miRNAs and human diseases from the database, thereby characterizing their relationship. The k-means clustering algorithm is employed to cluster most samples, selecting a certain number of samples from each cluster to balance the number of negative and positive samples. To further reduce feature dimension and time cost, probabilistic principal component analysis (PPCA) is applied to compress the feature vectors of the comprehensive similarity network of miRNA and disease, eliminating feature redundancy and then transforming the feature vectors into images. Subsequently, the prediction problem is converted into a 2-way 1-shot image problem and fed into the meta-learning model, with MAML serving as the meta-learner and CNN as the base learner. The results of a 5-fold cross-validation comparison on Dataset-H indicate that MAMLMDA surpasses several state-of-the-art methods, achieving the highest accuracy of 91%. Therefore, MAMLMDA offers a significant tool for understanding the pathogenesis of complex diseases at the miRNA level.