SARS-CoV-2 virus classification based on stacked sparse autoencoder.

Area of Science:

• Computational Biology
• Genomics
• Machine Learning

Background:

• The COVID-19 pandemic, caused by SARS-CoV-2, necessitates rapid viral classification for effective public health strategies.
• Deep learning has shown promise in various viral analysis tasks, including diagnosis and phylogenetics.

Purpose of the Study:

• To develop an efficient viral genome classifier for SARS-CoV-2 using deep neural networks.
• To evaluate the model's performance in taxonomic classification of viral genomes.

Main Methods:

• Utilized a stacked sparse autoencoder (SSAE) deep learning model.
• Employed k-mer image representations of complete genome sequences as input for the SSAE.
• Conducted four experiments for different taxonomic classification levels.

Main Results:

• Achieved high classification accuracy, ranging from 92% to 100% on the validation set.
• Demonstrated excellent performance on the test set, with accuracy between 98.9% and 100%.
• The model successfully classified SARS-CoV-2 samples not used during training, indicating adaptability.

Conclusions:

• The SSAE deep learning technique is highly effective for viral genome classification.
• The developed model shows potential for classifying emerging viruses and advancing genomic analysis.
• This approach highlights the applicability of deep learning in addressing critical challenges in virology.