Decoding the m⁶A epitranscriptomic landscape for biotechnological applications using a direct RNA sequencing approach

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Epitranscriptomic modifications, like N6-methyladenosine (m⁶A), regulate gene expression through RNA stability, splicing, and translation.
• Current computational methods for m⁶A detection in Nanopore direct RNA sequencing (DRS) data are limited by reliance on labeled data, leading to underestimation of modification sites.

Purpose of the Study:

• To introduce pum6a, an attention-based framework using positive and unlabeled multi-instance learning (MIL) to improve m⁶A detection from DRS data.
• To address challenges of incomplete labeling and missing read-level annotations in m⁶A site identification.

Main Methods:

• Pum6a integrates electrical signal features with base alignment data.
• A weighted Noisy-OR probability mechanism is employed for enhanced sensitivity and accuracy.
• The framework utilizes positive and unlabeled multi-instance learning (MIL) for robust annotation.

Main Results:

• Pum6a demonstrates superior performance in m⁶A detection across diverse cell lines and species, particularly in low-coverage regions.
• The tool outperforms existing methods without requiring extensive parameter tuning.
• Application to gastric cancer under hypoxia revealed distinct roles of FTO and ALKBH5 in m⁶A regulation and transcript stability.

Conclusions:

• Pum6a is a powerful tool for advancing the understanding of epitranscriptomic regulation.
• The findings provide key insights into m⁶A-mediated transcript stability in cancer.
• Pum6a has potential for biotechnological and therapeutic applications in epitranscriptomics.