Representation Learning Technique in DNA Sequence Classification for SARS-CoV-2 and Pathogenic Viruses

SARS-CoV-2 has demonstrated its ability to adapt and spread in various environments, making it a challenging target for identification and prediction. While current studies primarily use traditional statistical and biological analysis methods; however, implementing these methods is often time-consuming. To mitigate this problem, this study integrates representation learning and traditional machine learning techniques to enhance the classification performance of human virus DNA sequences. The proposed method focuses on the optimal combinations of k-mer sizes and the representation learning techniques with machine learning models to classify human virus DNA sequences. The integration of the FastText model, Random Forest classifier, and 3-mer achieved an impressive accuracy of 99.56%, showcasing the importance of sub-word information for genomics analysis. This approach underlines the potential for accurate and swift virus classification, essential for detecting crucial genomics variations and managing viral infections effectively.