PDMMDA: Prediction of MiRNA and Disease Association Based on Parallel Dimension Reduction Structure

MicroRNAs (miRNAs) play an important role in the development of organisms, and the interaction between miRNA and disease is also important for the control of diseases. Traditional biological assays to target disease-associated miRNAs are inefficient and expensive, so a more efficient and sensitive method is needed. We propose a new computational model to overcome this limitation, PDMMDA (Prediction of MiRNA and Disease Association), which uses a parallel dimensionality reduction structure to predict potential miRNA-disease associations. We combined miRNA and disease information from multiple databases to extract features of miRNA-disease association. Then, parallel deep autocoders and principal component analysis were used to extract the latent information for each miRNA-disease, and use a multilayer perceptron model for final prediction. Experimental results show that the optimal AUC value of PDMMDA is 0.9670, which has great potential for practical application.