A mapping-free natural language processing-based technique for sequence search in nanopore long-reads

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate gene expression analysis is crucial for dose estimation in radiation emergencies.
• Existing methods for targeted gene mapping can be computationally intensive and energy-consuming.
• There is a need for rapid, low-energy, in-situ sequencing analysis tools.

Purpose of the Study:

• To develop a sequence identification tool using natural language processing (NLP) techniques.
• To achieve a high negative predictive value (NPV) for gene expression analysis.
• To create a computationally inexpensive and energy-efficient alternative to classical mapping methods.

Main Methods:

• Exploration of various NLP models with different dictionary components and encoding lengths.
• Training and validation of NLP models on RNA sequencing data.
• Comparison of NLP model performance against classical mapping tools like minimap2.

Main Results:

• The optimal NLP configuration analyzed entire sequences with a 3-base pair word length.
• The best model achieved 98.29% balanced accuracy (BACC) and 99.25% NPV for the FDXR gene.
• Reduced dictionary size maintained high NPV (98.15% internally, 99.64% externally), with minimal impact on read count accuracy.

Conclusions:

• NLP-based mapping is a reliable replacement for classical methods in targeted transcript analysis of long nanopore reads.
• The developed model is adaptable for different transcript targets and long-read sequencing technologies.
• Classical text processing techniques show significant potential for nucleotide sequence analysis.