Deep sequence modelling for predicting COVID-19 mRNA vaccine degradation

Area of Science:

• Biotechnology
• Computational Biology
• Vaccinology

Background:

• The COVID-19 pandemic necessitates rapid vaccine development, with messenger RNA (mRNA) vaccines emerging as a promising candidate due to their rapid development, high potency, and potential for low-cost manufacturing.
• However, mRNA vaccine stability is a significant challenge, hindered by inefficient in vivo delivery, spontaneous degradation, and low protein expression levels.

Purpose of the Study:

• To design and implement a sequence deep model for predicting mRNA sequence degradation.
• To identify specific mRNA sequences responsible for degradation, thereby improving vaccine stability.

Main Methods:

• A deep learning model utilizing bidirectional Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) networks was developed.
• The model was trained on the Stanford COVID-19 mRNA vaccine dataset to predict five reactivity values per sequence position, indicating degradation likelihood and secondary structure.
• Features included numerical data and categorical data extracted from sequences, structure, and predicted loops, encoded using embedding layer learning.

Main Results:

• The LSTM model achieved a validation mean columnwise root mean square error (MCRMSE) of 0.125 with augmentation and codon encoding.
• Codon encoding demonstrated superior performance over Base encoding in terms of MCRMSE validation error.
• The model's ability to predict degradation and secondary structure offers insights into mRNA stability.

Conclusions:

• The developed deep learning model shows promise in predicting mRNA degradation, a critical factor for enhancing COVID-19 vaccine stability.
• The findings contribute to overcoming limitations in mRNA vaccine development, potentially accelerating the creation of more effective and stable vaccines.