RNA m6A detection using raw current signals and basecalling errors from Nanopore direct RNA sequencing reads

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Nanopore direct RNA sequencing (DRS) allows for RNA N6-methyladenosine (m6A) detection without additional laboratory steps.
• Current m6A detection methods from Nanopore DRS reads often rely on statistical signal features or basecalling errors, potentially overlooking crucial raw signal information.

Purpose of the Study:

• To introduce RedNano, a novel deep learning framework for enhanced m6A detection from Nanopore DRS data.
• To leverage both raw signal and basecalling error features for improved m6A identification.

Main Methods:

• RedNano employs residual networks to process distinct raw-signal and basecalling-error features from Nanopore DRS reads.
• The method was validated on synthesized, Arabidopsis, and human DRS datasets.

Main Results:

• RedNano consistently achieved superior performance, indicated by higher AUC and AUPR values across all tested datasets compared to existing methods.
• The model demonstrated robust cross-species validation capabilities.
• In an independent Populus trichocarpa dataset, RedNano exhibited significant improvements in AUC (3.8%-9.9%) and AUPR (5.5%-13.8%) over other methods.

Conclusions:

• RedNano offers a more effective approach for m6A detection using Nanopore DRS data by integrating comprehensive signal information.
• The method's robustness and superior performance highlight its potential for advancing epitranscriptomic research.