Development of deep learning models for RNA modification detection using Nanopore sequencing

Abstract

The expanding field of epitranscriptomics might rival the epigenome in the diversity of biological processes impacted and has the potential to answer some of the fundamental questions in RNA biology. A major roadblock in the epitranscriptomics field is the lack of transcriptome-wide methods to detect more than one RNA modification type at a time, identifying RNA modifications in individual molecules, and estimate modification stoichiometry accurately. The development of new computational techniques are a key driving force to address some of the limitations in the field and although multiple algorithmic applications for detecting and discovering RNA modifications have been developed, several challenges still exist to leverage the full potential of current experimentally produced datasets. Nanopore direct RNA sequencing (DRS) is a technology able to sequence single native RNA molecules and capture information about covalently bonded RNA modifications. This technology can be used to address some of the current limitations in the epitranscriptomic landscape. To address the current limitations, we developed CHEUI (CH3 (methylation) Estimation Using Ionic current), a two-stage deep learning method that detects N6-methyladenosine (m6A) and 5-methylcytidine (m5C) in individual RNA molecules from the same sample, as well as differential methylation between any two conditions. With CHEUI we developed the first deep learning algorithm able to detect several RNA modifications in individual reads, improving quantitative estimation of RNA modifications compared to previous methods, a major milestone in the field of DRS RNA modification detection. Moreover, we describe the protocol we used to train and test CHEUI, which can potentially be used to develop deep learning models to detect other RNA modifications. Finally, using CHEUI's unique capability to identify two modification types in the same sample we showed a non-random co-occurrence of m6A and m5C in mRNA transcripts in cell lines and tissues. We think CHEUI unlocks an unprecedented potential to study RNA modification configurations and discover new epitranscriptome functions.