Detection of m6A from direct RNA sequencing using a multiple instance learning framework.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• RNA modifications, like N6-methyladenosine (m6A) methylation, add complexity to transcriptome regulation.
• Nanopore direct RNA sequencing offers potential for detecting these modifications from raw signal data.
• Current methods lack read-level modification data for training machine learning models.

Purpose of the Study:

• To introduce m6Anet, a novel computational method for transcriptome-wide m6A detection.
• To address the challenge of missing single-molecule modification labels in training data.
• To enable accurate quantification of m6A modifications from direct RNA sequencing.

Main Methods:

• Development of m6Anet, a neural network utilizing a multiple instance learning framework.
• Leveraging site-level training data with missing read-level modification information.
• Application to Nanopore direct RNA sequencing data for m6A analysis.

Main Results:

• m6Anet demonstrates superior performance compared to existing computational methods.
• Achieves accuracy comparable to experimental approaches for m6A detection.
• Shows high generalization across different cell lines and species without model retraining.
• Accurately captures read-level stoichiometry to approximate m6A modification rates.

Conclusions:

• m6Anet provides a robust tool for identifying and quantifying m6A modifications.
• Enables comprehensive transcriptome-wide m6A analysis from a single direct RNA sequencing run.
• Facilitates deeper understanding of the role of m6A in gene regulation.