Detection of SARS-CoV-2 main variants of concerns using deep learning

The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has been a devastating effect on worldwide and is responsible for the coronavirus disease 2019 (COVID-19). The novel viral variants of the virus have been detected due to the high mutation rate of the virus. Although some SARS-CoV-2 variants can pose a significant risk for public health since they reduce the effect of the vaccine and medicine, some variants have a little effect on the health. The variants of SARS-CoV-2 may be classified in two parts, which are variants of concern and variants of interest. Variants of concern significantly decrease the effect of vaccines as well as increase the fatality. Five main SARS-CoV-2 variants of concerns, which are B.1.1.7 (Alpha), B.1.351 (Beta), P.1 (Gamma), B.1.617.2 (Delta), and B.1.1.529 (Omicron) are detected so far and it is crucial to identify these variants among the others. Variants of Interest, which are C.7 (Lambda) and B.1.621 (Mu) was excluded from the scope of this study. In this study, we propose a deep learning method using Convolutional Neural Network algorithm to detect main SARS-CoV-2 variants from human genome sequences. In addition, we also design primer sets to specific for each SARS-CoV-2 variant of concerns. In the proposed approach, first, features separating SARS-CoV-2 main variants of concerns are extracted using deep learning. Second, ensemble feature selection method is used to obtain considerable features. Machine learning classsifiers, which are k-nearest neighbor, support vector machine, random forest, and multilayer perceptron are performed to detect the variants. Then, candidate features obtained for each VOCs were analyzed in the Primer3Plus. At the end of study, we got results that is designed primer sets to detect SARS-COV-2 variant of concerns. Precise designing of primers and probes is critical to diagnose many variants of SARS-CoV-2. The proposed method exhibits admirable results and achieves an accuracy of 99.99% to classify variants. These results indicate that are considerable to improve diagnostic tools throughout Covid-19 pandemic and also future possible viral pandemics. However, the scarcity of available studies to identify the SARS-CoV-2 VOCs (especially B.1.1.529 variant) reveals the value of our study.