Convolutional Neural Network Applied to SARS-CoV-2 Sequence Classification.

Area of Science:

• Virology
• Genomics
• Computational Biology

Background:

• COVID-19, caused by SARS-CoV-2, is a global pandemic with significant public health implications.
• The emergence of highly transmissible variants like Omicron necessitates rapid identification and classification of viruses.
• Accurate taxonomic classification of SARS-CoV-2 from genomic data is crucial for strategic planning and disease control.

Purpose of the Study:

• To develop a high-accuracy deep learning technique for classifying viruses, specifically SARS-CoV-2, based on genome sequences.
• To overcome limitations in existing methods by not restricting genome sequence length.
• To provide a tool for accurate viral identification aiding in pandemic response.

Main Methods:

• Utilized a deep learning convolutional neural network (CNN) for genome sequence classification.
• Trained and tested the CNN on 1557 SARS-CoV-2 instances and 14,684 other viral sequences from NCBI and Virus-Host DB.
• Evaluated 48 different CNN architectures to determine optimal performance.

Main Results:

• The proposed CNN achieved 91.94 ± 2.62% accuracy in classifying viruses into their respective realms.
• Demonstrated 100% accuracy in classifying SARS-CoV-2 into the realm *Riboviria*.
• Accuracy increased for subsequent classifications (family, genus, subgenus), indicating high specificity.

Conclusions:

• The developed deep learning approach offers a viable and accurate method for classifying SARS-CoV-2 and other viruses using genomic sequences.
• This technique can support public health efforts in tracking, understanding, and managing viral outbreaks.
• The ability to handle variable genome lengths makes the method broadly applicable in virological research.