The Prediction of COVID-19 Virus Mutation Using Long Short-Term Memory

ABSTRACT

A worldwide epidemic has been caused by the new coronavirus (COVID-19). The high transmission rate of this pathogen requires early prediction and appropriate identification of mutations. Predicting this evolution will aid in the early detection of new strains and potentially facilitate the design of more effective antiviral therapies. However, SARS-CoV, MERS-CoV, and SARS-CoV2 (known as COVID-19) are all challenging to predict because of the virus's polymorphic nature, which enables it to adapt and survive across species, so there is a strong need for prediction to characterize mutations using their genetic information. A working method based on deep learning has been proposed to identify unknown sequences of pathogens to mitigate this problem. This study aims to predict virus mutations, especially codon mutations, for six types of coronavirus mutations (MERS-CoV, SARS-CoV-l, SARS-CoV-2, Alpha, Beta, Gamma). In this work, long-term memory is used for base prediction as an alignment-free technique. This algorithm is applied to several DNAs of coronavirus mutations where the k-mer technique is applied to segment the data to create a unique vocabulary. Then the TF-IDF is subsequently used for the identified virus sequences. The results showed that this technique's predictive accuracy on this data set reached 99%. It should be noted that this model was developed in Python using the Keras library, which is part of the Tensorflow library. © 2022 IEEE.