Prediction of RNA 5-Hydroxymethylcytosine Modifications Using Deep Learning

Abstract

It is becoming increasingly clear that RNA 5-hydroxymethylcytosine (5hmC), which plays an important role in several biological processes, is one of the most important objects of study in the field of RNA epigenetics. Biochemical experiments using various sequencing-based technologies are capable of achieving high-throughput identification of 5hmC, but current methods are labor-intensive, costly, and time-consuming. There is an imperative need to develop more efficient and robust computational methods to replace, or at least complement, such high-throughput methods. Although one such machine learning-based model to achieve this has already been developed, its performance is limited. In this study, we developed iRhm5CNN, an efficient and reliable computational predictive model for the identification of RNA 5hmC sites. Our model is based on a convolution neural network (CNN) that extracts the most reliable feature from the RNA sequence inevitably. The results of our experiments show significant outperformance across all evaluation metrics of our proposed architecture when compared to the only existing state of the art computational model in all the evaluation metrics. The proposed model can be accessed for free at http://nsclbio.jbnu.ac.kr/tools/iRhm5CNN/.

Keywords

• Post-transcriptional modification
• RNA 5-hydroxymethylcytosine
• sequence analysis
• convolutional neural network
• deep learning