Detecting m6A RNA modification from nanopore sequencing using a semi-supervised learning framework.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Direct nanopore RNA sequencing offers a method for detecting RNA base modifications like N6-methyladenosine (m6A) methylation by analyzing electrical current signals.
• A significant hurdle in this field is the limited availability of comprehensive training datasets for methylation detection.

Approach:

• We developed Xron, a hybrid encoder-decoder framework, for direct m6A methylation basecalling.
• Xron is trained in two stages: first, using synthetic RNA data generated via in silico cross-linking for diverse modification combinations, and second, fine-tuning on immunoprecipitation-based experimental data with label smoothing.
• This approach trains an end-to-end neural network basecaller.

Key Points:

• The Xron basecaller directly distinguishes methylated bases from raw nanopore sequencing signals.
• In silico cross-linking enhances the diversity of synthetic training data for improved model performance.
• Fine-tuning with label smoothing on experimental data further refines the basecaller's accuracy.

Conclusions:

• The trained Xron basecaller demonstrates superior performance compared to existing methods in both read-level and site-level m6A detection.
• Xron is a standalone, end-to-end solution for direct m6A detection from nanopore RNA sequencing data.
• This technology facilitates de novo methylome assembly, advancing the study of RNA modifications.