Multifractal analysis and support vector machine for the classification of coronaviruses and SARS-CoV-2 variants

Abstract

This study presents a novel approach for the classification of coronavirus species and variants of SARS-CoV-2 using Chaos Game Representation (CGR) and 2D Multifractal Detrended Fluctuation Analysis (2D MF-DFA). By extracting fractal parameters from CGR images, we constructed a state space that effectively distinguishes different species and variants. Our method achieved [Formula: see text] accuracy in species classification, with a notable [Formula: see text] accuracy for SARS-CoV-2 variants despite their genetic similarities. Using a Support Vector Machine (SVM) as a classifier further enhanced the performance. This approach, which requires fewer steps than most existing methods, offers an efficient and effective tool for viral classification, with implications for bioinformatics, public health, and vaccine development.

Keywords: Coronaviridae; CGR; Fractal; RNA; SVM; Viral.

Fig. 1

Graphical summary of workflows.

Fig. 2

Chaos game representation for coronavirus species: HCoV-HKU1, HCoV-OC43, HCoV-NL63, HCoV-229E, HCoV-MERS, SARS-CoV-2. We used the samples identified as reference sequences in NCBI.

Fig. 3

Power-law multifractal nature of coronavirus species CGR image. Some constants are subtracted to make the contrast between the different curves clearer in graphics of F(q) vs. q. The straight lines are the best-fit lines whose slopes are shown in the legend.

Fig. 4

Multifractal spectrum of the coronavirus species (above) and the variants of SARS-CoV-2 (below).

Fig. 5

Scatter plots of fractal parameters.

Fig. 6

State space constructed using fractal parameters (h(2), , ) for SARS-CoV-2 variants. The apparent mixing of certain variants may reflect their evolutionary proximity or similarities in genomic features.

Fig. 7

Left: The accuracy of the six coronavirus species for the selected combinations with increasing K. Right: The accuracy of the five SARS-CoV-2 species variants for the selected combinations with increasing K.

Fig. 8

Confusion Matrix. Each row represents the actual class, and each column represents the predicted class. The diagonal elements indicate correctly classified samples. Classes: 0 (Alpha variant), 1 (Beta variant), 2 (Delta variant), 3 (Gamma variant), 4 (Omicron variant).