The mRNA COVID-19 vaccine plays a key role in the fight against COVID-19. Because of its rapid production ability and promising results in a number of clinical studies, vaccines and treatments based on mRNA technology are getting more and more attention.
However, due to the thermal instability of mRNA, it is vulnerable to chemical degradation, which is also a major challenge for mRNA-based vaccines or treatments. Strict conditions are required for the production, storage, and transportation of mRNA vaccines. In order to make mRNA vaccines
Recently, the Sun Qing team of Texas A&M University published a research paper entitled “RNAdegformer: accurate prediction of mRNA degradation at nucleotideAAI resolution with deep learning” in the journal Briefings in Bioinformatics.
The research team used deep learning technology to create an effective and interpretable model architecture, RNAdegformer, which can predict RNA degradation more accurately than previous best methods, such as the Degscore model, the RNA folding algorithm, and other machine learning models.
Professor Sun Qing said that the chemical degradation reaction caused by the inherent thermal instability of mRNA hinders the global distribution of mRNA vaccines, so the study attempted to understand and predict mRNA degradation.
The research team used artificial neural networks to power RNAdegformer, a deep learning-based model that can extract data and use these insights for prediction, in order to address the issue of mRNA degradation.
RNAdegformer makes use of the biophysical characteristics of RNA secondary structure and base pairing probability, and uses self-attention and convolution to process RNA sequences. These two deep learning techniques have been proven to play a dominant role in the fields of computer vision and natural language processing.
RNAdegformer is superior to the previous best method in predicting degradation characteristics at the nucleotide level, and RNAdegformer can predict each nucleotide in COVID-19 mRNA vaccine. Compared with the previous best methods, the correlation between RNAdegformer prediction and the in vitro half-life of RNA was also improved.
In addition, the study shows how direct visualization of self-attention images can contribute to informed decision-making. Attention tries to show how the model uses input information to “think”, which is helpful to informed decision-making based on model prediction. In addition, this model also reveals the basic characteristics that determine the degradation rate of mRNA. The team worked with Rhiju Das, an associate professor of biochemistry at Stanford University, whose high-quality mRNA degradation data were the starting point for the study.
Professor Sun Qing said that through this study, we hope to be able to use our model to design a more stable mRNA vaccine, so that mRNA therapy can be used more fairly and widely.