Classification of Highly Divergent Viruses from DNA/RNA Sequence Using Transformer-Based Models.

Area of Science:

• Computational Virology
• Bioinformatics
• Genomics

Background:

• Viruses pose significant global health risks, causing infections and increasing cancer risk.
• Emerging viral threats like COVID-19 and influenza highlight the need for advanced detection methods.
• Traditional virology struggles with novel or divergent viral genomes, necessitating new analytical approaches.

Purpose of the Study:

• To develop a deep learning system for accurate identification of viral DNA/RNA sequences.
• To address the limitations of traditional methods in detecting novel and divergent viruses.
• To improve the differentiation of viral pathogens, variants, and strains.

Main Methods:

• Utilized advanced deep learning, specifically a custom BERT architecture for DNA analysis.
• Employed nucleotide sequences from NCBI GenBank and a BERT tokenizer for feature extraction.
• Incorporated synthetic data generation for viruses with limited sample sizes.
• Developed a two-component system: unsupervised next-codon learning and a genotype-phenotype classifier.

Main Results:

• Achieved a high accuracy of 97.69% in identifying viral sequences.
• Demonstrated the system's capability to identify dozens of different viruses.
• Successfully extracted domain-specific features from nucleotide sequences using deep learning.

Conclusions:

• The proposed deep learning system offers a robust and accurate method for viral sequence identification.
• This computational virology approach enhances pathogen detection, aiding in public health surveillance.
• The system effectively handles challenges posed by viral genome variability and limited data.