Messenger RNA (mRNA)-based vaccines have emerged as rapid and versatile candidates to quickly respond to virus pandemics such as the coronavirus. Pfizer and Moderna have successfully developed effective mRNA vaccines to fight COVID-19. However, one key limitation of mRNA based vaccines and therapeutics is the inherent instability of mRNA molecules. Currently, little is known on where in the backbone of a given RNA is most likely to degrade. Therefore, understanding and predicting mRNA degradation is crucial. In this study, we show how biophysics and deep learning can be combined to learn from experimental data to accurately predict mRNA degradation under different conditions at nucleotide resolution. Our deep learning models achieve up to 0.68 R2 score on mRNA degradation prediction of each nucleotide. Additionally, we demonstrate these predictions are strongly correlated with half-lives of unseen mRNA sequences with up to -0.657 Pearson's r and can thus be used to estimate half-lives of mRNA sequences. Finally, we propose how our deep learning models can be used to stablize mRNA sequences via computational design.

Accurate prediction of mRNA degradation at nucleotide resolution with deep learning

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.