COVID-19 mRNA Vaccine Degradation Prediction using Regularized LSTM Model

Imran, S. A.; Islam, M. T.; Shahnaz, C.; Imam, O. T.; Haque, M.; Ieee,; Islam, Md T.

Due to the advantages of mRNA vaccines such as potency, safety, and production feasibility, recent researches in vaccinology has seen strong focus in mRNA vaccines. As leading researches involving COVID-19 mRNA vaccine candidates are being carried out, the challenge of overcoming the stability tradeoff of mRNA vaccines stand between the production and effective mass distribution stages. With the help of the OpenVaccine RNA database with degradation rate measurements provided by Stanford researchers, we developed an artificial recurrent neural network model to help bioinformatics researchers identify whether and where mRNAs might be unstable and prone to degrade under certain incubation measures. For this purpose, we've prepared a regularized LSTM model which minimizes mean columnwise root mean squared error for several degradation rates. We've found that recurrent algorithms perform better than tree-based algorithms.