Genomic language model mitigates chimera artifacts in nanopore direct RNA sequencing

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Chimera artifacts in nanopore direct RNA sequencing (dRNA-seq) reduce accuracy in transcript annotation and gene fusion detection.
• Existing basecalling models cannot identify or fix these artifacts, limiting dRNA-seq utility.

Purpose of the Study:

• To develop a method for identifying and removing adapter sequences from dRNA-seq long reads.
• To enhance the reliability and accuracy of nanopore dRNA-seq for transcriptomics.

Main Methods:

• Developed DeepChopper, a genomic language model for adapter sequence identification and removal.
• DeepChopper operates on base-called dRNA-seq reads, independent of raw signal or alignment data.
• The model precisely removes adapter-bridged artifacts with single-base resolution.

Main Results:

• DeepChopper effectively identifies and removes adapter sequences from dRNA-seq long reads.
• The method eliminates adapter-bridged artifacts without requiring raw signal or alignment information.
• Demonstrated enhancement in the accuracy of downstream transcriptomics analyses.

Conclusions:

• DeepChopper significantly improves the accuracy of nanopore dRNA-seq data.
• The tool enhances transcript annotation and gene fusion detection capabilities.
• DeepChopper establishes nanopore dRNA-seq as a more robust tool for transcriptomics research.