A Novel Feature Fusion Method Based on RSCU and K-mer to Classify the SARS-Cov-2

The SARS-Cov-2 virus exhibits a high mutation rate, which makes the prediction and classification of its genetic evolution and variation trends highly significant. Accurate classification methods not only contribute to epidemiological studies of the virus, but also play a crucial role in vaccine development and antiviral drug discovery. This study aims to systematically evaluate the accuracy and generalization capability of the RSCU (Relative Synonymous Codon Usage) and K-mer encoding techniques in the classification of the SARS-CoV-2 genome. We extracted genomic data from two major SARS-CoV-2 variants, Alpha and Beta, and applied the Support Vector Machine (SVM) classification algorithm to train the data and assess the impact of different feature encoding methods on classification performance. Furthermore, we introduce a novel multi-feature fusion method, KRSCU, which combines the sequence position information from K-mer with the synonymous codon compositions from RSCU. This method effectively captures subtle differences in genomic data, significantly improving both the accuracy and generalization capability of the classification model. Experimental results demonstrate that the KRSCU method outperforms traditional single-feature encoding approaches in SARS-CoV-2 subtype classification tasks. Our research offers new insights into genomic data analysis, with potential applications in viral mutation monitoring.