Mixed-Weight Neural Bagging for Detecting m⁶A Modifications in SARS-CoV-2 RNA Sequencing

Area of Science:

• Virology
• Molecular Biology
• Bioinformatics

Background:

• N6-methyladenosine (m6A) is a common RNA modification influencing viral gene mutation and protein structure in SARS-CoV-2.
• Nanopore direct RNA sequencing (DRS) preserves m6A signatures but lacks robust detection methods due to insufficient data.
• Identifying m6A modifications is critical for understanding SARS-CoV-2 infection mechanisms and protein expression.

Purpose of the Study:

• To develop and validate a precise method for identifying m6A RNA modifications directly from Nanopore DRS data.
• To apply this method for the identification of m6A modifications in the SARS-CoV-2 genome.

Main Methods:

• A methodology was developed involving mapping and feature extraction from DRS data.
• An ensemble machine learning model, mixed-weight neural bagging (MWNB), was introduced for m6A detection.
• The MWNB model was trained using synthetic DRS data of modified and unmodified RNA.

Main Results:

• The MWNB model achieved high classification accuracy (97.85%) and AUC (0.9968) in identifying m6A modifications.
• Application of the MWNB model to a COVID-19 dataset showed strong correlations with existing biomedical findings.
• The study successfully identified m6A modifications on the SARS-CoV-2 RNA.

Conclusions:

• The developed strategy enables accurate prediction of m6A modifications using DRS data.
• This work provides a crucial tool for the identification of m6A modifications in SARS-CoV-2.
• The findings contribute to a deeper understanding of viral RNA modifications and their impact on infection.