RMNet: An RNA m6A Cross-species Methylation Detection Method for Nanopore Sequencing.

Area of Science:

• Epitranscriptomics
• Bioinformatics
• Genomics

Background:

• N6-methyladenosine (m6A) is a crucial RNA modification impacting gene expression.
• Current m6A detection methods are often slow, labor-intensive, or lack broad applicability.
• Existing machine learning models for m6A detection frequently fail to generalize across different species.

Purpose of the Study:

• To develop RMNet, a novel and robust method for cross-species m6A detection.
• To leverage nanopore sequencing data for efficient and accurate m6A site identification.
• To create a single machine learning model applicable to diverse organisms.

Main Methods:

• RMNet integrates signal and alignment features from nanopore sequencing data.
• The model utilizes Conformer and RNN architectures.
• Contrastive learning is employed to improve the distinction between m6A and non-m6A sites.

Main Results:

• RMNet achieved high accuracy: 99.7% (synthesized RNA), 78.8% (Arabidopsis), and 88.9% (human).
• The model demonstrated superior performance across six metrics, including AUC and AUPR, compared to existing methods.
• RMNet exhibited robust cross-species generalization capabilities with a single set of model weights.

Conclusions:

• RMNet offers a unified, efficient, and sensitive approach for m6A detection across species.
• The method advances epitranscriptomics research with potential applications in precision medicine and agricultural science.
• Identified limitations include potential challenges with thymine-rich k-mers in human datasets due to secondary structures.