Detecting m6A RNA modification from nanopore sequencing using a semisupervised learning framework

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Direct nanopore sequencing of RNA can identify posttranscriptional base modifications like N6-methyladenosine (m6A) via electrical signals.
• A significant hurdle is the limited availability of training data for methylation detection.

Purpose of the Study:

• To develop Xron, a novel hybrid encoder-decoder framework for direct methylation-distinguishing basecalling.
• To address the challenge of insufficient training data in RNA methylation detection.

Main Methods:

• Utilized in silico cross-linking to generate diverse RNA modification combinations for synthetic data.
• Employed a two-step training process: initial training on synthetic data, followed by fine-tuning on immunoprecipitation (IP)-based experimental data with label smoothing.
• Developed an end-to-end neural network basecaller within the Xron framework.

Main Results:

• The trained Xron basecaller demonstrated superior performance compared to existing methods in both read-level and site-level methylation prediction.
• Successfully enabled direct detection of methylated bases from raw nanopore sequencing signals.

Conclusions:

• Xron is a capable, standalone tool for end-to-end m6A detection directly from sequencing data.
• Facilitates de novo methylome assembly by directly identifying methylation sites.