Abstract The most prevalent post-transcriptional RNA modification, pseudouridine (Ψ), also known as the fifth ribonucleoside, is widespread in rRNAs, tRNAs, snRNAs, snoRNAs and mRNAs. Pseudouridines in RNAs are implicated in many aspects of post-transcriptional regulation, such as the maintenance of translation fidelity, control of RNA stability and stabilization of RNA structure. However, our understanding of the functions, mechanisms as well as precise distribution of pseudourdines (especially in mRNAs) still remains largely unclear. Though thousands of RNA pseudouridylation sites have been identified by high-throughput experimental techniques recently, the landscape of pseudouridines across the whole transcriptome has not yet been fully delineated. In this study, we present a highly effective model, called PULSE ( P seudo U ridy L ation S ites E stimator), to predict novel Ψ sites from large-scale profiling data of pseudouridines and characterize the contextual sequence features of pseudouridylation. PULSE employs a deep learning framework, called convolutional neural network (CNN), which has been successfully and widely used for sequence pattern discovery in the literature. Our extensive validation tests demonstrated that PULSE can outperform conventional learning models and achieve high prediction accuracy, thus enabling us to further characterize the transcriptome-wide landscape of pseudouridine sites. Overall, PULSE can provide a useful tool to further investigate the functional roles of pseudouridylation in post-transcriptional regulation.

Characterizing RNA Pseudouridylation by Convolutional Neural Networks