CLRSMMA: A prediction framework for small molecule drug-miRNA associations using convolutional neural network and logistic regression algorithm models

MicroRNA (miRNA) holds a crucial position in a wide range of biological processes and human diseases and is seen as a potential therapeutic target for small molecule drugs. Given that verifying the relationships between small molecule drugs and miRNAs necessitates biological experiments which consume a large amount of time and money, it is pressing to create new computational methods to forecast new small molecule drug - miRNA associations. The use of deep learning-related methods to predict small molecule drug-miRNA associations overcomes the shortcomings of traditional biological experiments to identify small molecule drug-related miRNAs, such as high economic cost, large blindness, and long experimental cycle. In this paper, we proposed a CLRSMMA method based on convolutional neural network and logistic regression algorithm to predict small molecule drug-miRNA associations. First, a high-dimensional feature matrix was formed by combining the similarity between small molecule drugs, the similarity between miRNAs, and the known associations of small molecule drugs with miRNAs. Secondly, a convolutional neural network was used to perform feature dimensionality reduction on the fusion matrix to obtain the potential feature representation of each small molecule drug-miRNA pair. Finally, the extracted features were sent to the logistic regression algorithm to infer the potential association between small molecule drugs and miRNAs. We conducted a five-fold cross-validation experiment on an independent dataset, and the results showed that CLRSMMA outperformed various state-of-the-art methods. Furthermore, in both case studies, a large number of associations predicted by CLRSMMA were confirmed with publicly accessible literature. Therefore, CLRSMMA is considered a reliable method for small molecule drug-miRNA association prediction.