CNNPSP: Pseudouridine sites prediction based on deep learning

Abstract

Pseudouridine (ψ) is a kind of RNA modification, which is formed at specific site of RNA sequence due to the catalytic action of Pseudouridine synthase in the process of gene transcription. It is the most prevalent RNA modification found so far, and plays a vital role in normal biological functions. Several computational methods have been proposed to predict Pseudouridine sites, but these methods still do not achieve high accuracy. At present, deep learning has become a popular field of machine learning, and convolutional neural network (CNN) is one widely used algorithm. CNN can automatically dig into the hidden features of data and make accurate predictions, so a new algorithm based on CNN was proposed for extracting the features from RNA sequences with and without ψ sites. And a predictor called CNNPSP was developed to predict ψ sites in RNAs across three species (H. sapiens, S. cerevisiae and M. musculus). Both the rigorous jackknife test and independent test indicated that the new predictor outperformed the existing methods in this task.