N-Gram-Codon and Recurrent Neural Network (RNN) to Update Pfizer-BioNTech mRNA Vaccine

Abstract

In the fight against SARS-CoV-2, Pfizer BioNTech based on synthetic messenger RNA (mRNA) proved to be quicker and more effective even with a small dose of micrograms per injection. Unfortunately, such a vaccine requires very low temperatures to prevent degradation of mRNA. In this paper, we have developed three new models of recurrent neural network (1- simple LSTM 2-BDLSTM 3-BERT) using n-gram-codon technique for the codification of mRNA. The primary aim is to analyse the mRNA sequence and predict the stability/reactivity rates at various codon positions. The results of the predictions will be presented in the form of recommendations to support laboratories in updating Pfizer's BioNTech vaccine. The obtained results were validated by the Stanford OpenVaccine dataset and the evaluation measures recall, precision, f1-score, accuracy and loss.