A convolution neural network-based computational model to identify the occurrence sites of various RNA modifications by fusing varied features

Abstract

RNA modification occurs in both prokaryotic and eukaryotic genomes, which is considered one of the major RNA properties. RNA modifications are the main portions of the regulatory landscape found in genes, which contain several bioprocesses at the post-transcriptional level. Therefore, the identification of RNA modifications residue information is essential for determining their molecular functions and their relevant mechanisms. Although the wet lab experimental works for identification of RNA modification sites have produced satisfactory results, these experimental-based approaches are highly labor-intensive and precious. So, it is indispensable to establish a novel and robust computational approach for the prediction of RNA modification sites. To solve these issues, an intelligent computational predictor called “iRNA-Mod-CNN”, using deep learning hypotheses is developed to identify RNA modification sites. First, the biological sequences are encoded by implementing the one-hot encoding method. Then encoded feature vector is provided to the convolution neural network (CNN) model in order to discern the conceal information. Further, k-Gram feature space is amalgamated with CNN feature space. The computational predictor “iRNA-Mod-CNN” showed significant improvement over the existing methods, producing 99.56%, 92.39%, and 86.66% of accuracies on m1A, m6A, and m5C benchmark datasets, respectively.