Ab initio detection of multiple epitranscriptomic modifications from Oxford nanopore technology direct RNA sequencing...

Area of Science:

• Bioinformatics
• Molecular Biology
• Genomics

Background:

• Direct RNA sequencing offers a promising avenue for studying the eukaryotic epitranscriptome.
• Current bioinformatics tools struggle with simultaneous detection of multiple modified bases, requiring specialized, modification-specific approaches.
• Existing methods are often modification-unaware or necessitate multiple, complex learning steps.

Purpose of the Study:

• To introduce NanoSpeech, a novel modification-aware basecaller for simultaneous detection of multiple RNA bases.
• To present NanoListener, a strategy for generating robust training datasets for basecalling.
• To develop a new generation of Oxford Nanopore Technologies (ONT) basecallers independent of specific ONT chemistry.

Main Methods:

• Developed NanoSpeech, a transformer-based basecaller for ab initio detection of multiple modified bases.
• Implemented NanoListener using a simulated randomers strategy to create robust training datasets.
• Designed a single basecalling model with an expanded vocabulary capable of identifying both unmodified and modified bases.

Main Results:

• NanoSpeech enables simultaneous detection of multiple modified bases from direct RNA sequencing data.
• NanoListener provides robust training datasets, enhancing the accuracy of basecalling models.
• A single, trained model can accurately basecall both unmodified and modified RNA bases, regardless of ONT chemistry.

Conclusions:

• NanoSpeech and NanoListener significantly advance the analysis of the eukaryotic epitranscriptome.
• These tools overcome limitations of current modification-unaware bioinformatics software.
• The developed approach facilitates accurate and simultaneous identification of diverse RNA modifications.