PAPET: a collection of performant algorithms to identify 5-methyl cytosine from PacBio SequelII data

Abstract

BackgroundCytosine followed by guanine (CpG) rich genomic regions are known important regulatory sequences among vertebrates. Their cytosines can be subjected to a variety of chemical modifications, one of which being 5-methylcytosine (5mC), usually coined CpG methylation. CpG methylation has been demonstrated to have a deep impact on the nearby regulated genes. The advent of single molecule real time sequencing methods have proven to be powerful enablers in this field of research because of the possibility to detect DNA chemical modification directly from the raw sequencing data. In this perspective, several 5mC detection methods have been proposed. ResultsIn this work we present PAPET - PacBio Prediction of Epigenetic Toolkit - a computational method to detect 5mC directly from PacBio SequelII raw sequencing data. We characterized the 5mC signatures in the raw data and proposed a framework to model them. We also assessed the effect of the DNA sequence alone on the signal and propose a normalization method to leverage this effect. From this, we designed a probabilistic approach to predict the presence of cytosine methylation from the sequencing kinetics and benchmarked it. ConclusionsWe apply this framework to predict CpG methylation from SequelII data and demonstrate that the classifiers compare equally with PacBios prediction method counterparts in terms of AUC and achieve very high specificity and precision.