Article: A comparative study on the classification of SARS-CoV-2 variants from biosequence images using pre-trained deep learning models Journal: International Journal of Bioinformatics Research and Applications (IJBRA) 2025 Vol.21 No.1 pp.102 - 119 Abstract: Coronavirus disease has raised serious health concern across the globe. Identification of severe acute respiratory syndrome corona virus 2 (SARS-CoV-2) variants are indeed a concern in controlling its spread. SARS-CoV-2 variants are classified based on the variation in its genomic sequences. Alpha, beta, delta, gamma and omicron were reported as the most deleterious variants. Genome sequence can be represented uniquely using chaos game representation (CGR) images. A large-scale genome sequence dataset, belonging to the five categories of these variant were retrieved from GISAID. An attempt was made to compile benchmark CGR images of 25,000 SARS-CoV-2 variants genomic sequences. The present study aims to compare the performance of different pre-trained deep learning models in classifying SARS-CoV-2 variants from its CGR images. VGG16, VGG19, ResNet50, InceptionV3, Xception, InceptionResNetV2 and MobileNetV2 were the models used for the study. SARS-CoV-2 variant detection was found effective with VGG19 with an accuracy of 94%. Data augmentation techniques were also applied on the CGR images of biosequences and it was found that data augmentation methods decreased the accuracy of different transfer learning models. Inderscience Publishers - linking academia, business and industry through research

Title: A comparative study on the classification of SARS-CoV-2 variants from biosequence images using pre-trained deep learning models

Authors: K. Shahina; C.L. Biji; Achuthsankar S. Nair

Addresses: Department of Computational Biology and Bioinformatics, University of Kerala, Thiruvananthapuram, Kerala, India ' Department of Analytics, School of Computer Science and Technology, Vellore Institute of Technology, Vellore, Tamil Nadu, India ' Department of Computational Biology and Bioinformatics, University of Kerala, Thiruvananthapuram, Kerala, India

Abstract: Coronavirus disease has raised serious health concern across the globe. Identification of severe acute respiratory syndrome corona virus 2 (SARS-CoV-2) variants are indeed a concern in controlling its spread. SARS-CoV-2 variants are classified based on the variation in its genomic sequences. Alpha, beta, delta, gamma and omicron were reported as the most deleterious variants. Genome sequence can be represented uniquely using chaos game representation (CGR) images. A large-scale genome sequence dataset, belonging to the five categories of these variant were retrieved from GISAID. An attempt was made to compile benchmark CGR images of 25,000 SARS-CoV-2 variants genomic sequences. The present study aims to compare the performance of different pre-trained deep learning models in classifying SARS-CoV-2 variants from its CGR images. VGG16, VGG19, ResNet50, InceptionV3, Xception, InceptionResNetV2 and MobileNetV2 were the models used for the study. SARS-CoV-2 variant detection was found effective with VGG19 with an accuracy of 94%. Data augmentation techniques were also applied on the CGR images of biosequences and it was found that data augmentation methods decreased the accuracy of different transfer learning models.

Keywords: genome sequence; deep learning; SARS-CoV-2 variants; chaos game representation; transfer learning; classification; COVID-19.

DOI: 10.1504/IJBRA.2025.144026

International Journal of Bioinformatics Research and Applications, 2025 Vol.21 No.1, pp.102 - 119

Received: 08 Dec 2023
Accepted: 19 Feb 2024
Published online: 21 Jan 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: A comparative study on the classification of SARS-CoV-2 variants from biosequence images using pre-trained deep learning models

Keep up-to-date