Nm-Nano: A Machine Learning Framework for Transcriptome-Wide Single Molecule Mapping of 2´-O-Methylation (Nm) Sites in Nanopore Direct RNA Sequencing Datasets

Abstract

Nm (2′-O-methylation) is one of the most abundant modifications of mRNAs and non-coding RNAs occurring when a methyl group (–CH3) is added to the 2′ hydroxyl (–OH) of the ribose moiety. This modification can appear on any nucleotide (base) regardless of the type of nitrogenous base, because each ribose sugar has a hydroxyl group and so 2′-O-methyl ribose can occur on any base. Nm modification has a great contribution in many biological processes such as the normal functioning of tRNA, the protection of mRNA against degradation by DXO, and the biogenesis and specificity of rRNA. Recently, the single-molecule sequencing techniques for long reads of RNA sequences data offered by Oxford Nanopore technologies have enabled the direct detection of RNA modifications on the molecule that is being sequenced, but to our knowledge there was only one research attempt that applied this technology to predict the stoichiometry of Nm-modified sites in RNA sequence of yeast cells. To this end, in this paper, we extend this research direction by proposing a bio-computational framework, Nm-Nano for predicting Nm sites in Nanopore direct RNA sequencing reads of human cell lines. Nm-Nano framework integrates two supervised machine learning models for predicting Nm sites in Nanopore sequencing data, namely Xgboost and Random Forest (RF). Each model is trained with set of features that are extracted from the raw signal generated by the Oxford Nanopore MinION device, as well as the corresponding basecalled k-mer resulting from inferring the RNA sequence reads from the generated Nanopore signals. The results on two benchmark data sets generated from RNA Nanopore sequencing data of Hela and Hek293 cell lines show a great performance of Nm-Nano. In independent validation testing, Nm-Nano has been able to identify Nm sites with a high accuracy of 93% and 88% using Xgboost and RF models respectively by training each model with Hela benchmark dataset and testing it for identifying Nm sites on Hek293 benchmark dataset. Thus, Nm-Nano outperforms the Nm sites predictors existing in the literature (not relying on Nanopore technology) that were only limited to predict Nm sites on short reads of RNA sequences and unable to predict Nm sites on long RNA sequence reads. By deploying Nm-Nano to predict Nm sites in Hela cell line, it was revealed that a total of 196 genes was identified to have the most abundance of Nm modification among all other genes that have been modified by Nm in this cell line. Similarly, deploying Nm-Nano to predict Nm sites in Hek393 cell line revealed that a total of 196 genes line was identified to have the most abundance of Nm modification among all other genes that have been modified by Nm in this cell line. According to this, a significant enrichment of a wide range of functional processes like high confidences (adjusted p-val < 0.05) enriched ontologies that were more representative of Nm modification role in immune response and cellular homeostasis were revealed in Hela cell line, and "MHC class 1 protein complex", "mitotic spindle assembly", "response to glucocorticoid", and "nucleocytoplasmic transport" were revealed in Hek293 cell line. The source code of Nm-Nano can be freely accessed https://github.com/Janga-Lab/Nm-Nano.