A novel Lucas series-based adaptive numerical representation of amino acid sequences for prediction of COVID-19 variants

Abstract

The COVID-19 disease, caused by the SARS-CoV-2 virus, quickly spread worldwide and developed into a pandemic. Given the virus's rapid spread and high transmissibility, early-stage diagnosis of the disease is crucial. Today, vaccines have begun to be produced to control the rate of increase and spread of the virus. However, the constant appearance of virus replications can lead to the emergence of mutant viruses and prevent the formation of potential antibodies. In addition, various chemical processes are used to identify virus variants, which increase the need for laboratories and make the process costly. Due to these disadvantages, it is recommended to use computational-based approaches to quickly identify variants of the SARS-CoV-2 virus. To analyze proteins with computational-based approaches, protein sequences need to be converted to the numerical representations. In this study, using a computational-based approach, a newly developed protein mapping approach called LucasProtein (LUCPROT) was introduced and applied to predict variants of the SARS-CoV-2 virus. The study consists of four stages: obtaining protein sequence data, converting protein sequences to the numerical representations, determining the classification models, and performing the prediction. Based on the application results, the proposed protein mapping method successfully predicted SARS-CoV-2 virus variants, achieving an accuracy of 86.5% and an ROC-AUC of 0.75. The findings regarding the proposed mapping method and the performance of the COVID-19 variant prediction demonstrate that the framework can be effectively implemented.