Forecasting Mutated Genome Sequences of Human Coronaviruses by Unveiling Genomic Mutations Leveraging the Seq2Seq Approach

Since viruses mutate rapidly, an outbreak of SARS-CoV-2, otherwise known as Covid-19, that originated a few years ago, has persisted in humans for nearly four years and is showing its rapid unpredictable mutation nature continuously. Several human coronavirus variants have emerged, including SARS-CoV in 2002, MERS-CoV in 2012, and the most recent, SARS-CoV-2 in 2019. This virus caused a major global outbreak, leading to over 700 million infections and more than 7 million deaths worldwide by March 2024. The continuous evolution of new SARS-CoV-2 strains has created a major challenge, and researchers are working hard to find effective solutions to tackle the problem. The most recent variant JN.1 has caused serious concern among public health officials due to its rapid spread and potential impact on vaccine efficacy and disease severity. To address this, the study uses a new approach with a Seq2Seq model to predict the genomic sequences of future Human Coronavirus variants early. The model learns the patterns of genomic mutations by analyzing the genome sequences of past Human Coronavirus variants. The genome sequences were arranged chronologically according to the date of their original discovery. A sliced approach for genome sequence is used for better learning of the model. The model produced a well fit validation loss curve with an average validation accuracy of 0.93. This research work demonstrates the possibility to predict the mutated genome sequence by detecting the chronological mutation patterns through natural language processing models such as the Seq2Seq model. In order to tackle the issue of rapid mutation of SARS-CoV-2, the proposed model would work as a useful method for predicting the genomic sequences of upcoming variations and obtaining crucial genomic insights and structures of the upcoming variants. The code, data and results are available in the GitHub repository.