CodonBERT large language model for mRNA vaccines

Area of Science:

• Biotechnology
• Computational Biology
• Genomics

Background:

• Messenger RNA (mRNA) technology is increasingly vital for vaccines and therapeutics.
• Optimizing mRNA sequences is crucial for enhancing expression, stability, and immunogenicity.
• A vast number of mRNA sequences can encode even small proteins, necessitating efficient design tools.

Purpose of the Study:

• To develop an advanced large language model (LLM) for optimizing mRNA sequences.
• To create a tool that captures complex biological concepts within mRNA sequences.
• To improve the prediction of mRNA properties for therapeutic and vaccine development.

Main Methods:

• Developed CodonBERT, an LLM that utilizes codons as input for enhanced representation learning.
• Trained CodonBERT on over 10 million diverse mRNA sequences from various organisms.
• Evaluated CodonBERT's performance on predicting mRNA properties and compared it to existing methods.

Main Results:

• CodonBERT effectively learns representations from mRNA sequences using codons as input.
• The model demonstrates the ability to capture significant biological insights from mRNA data.
• CodonBERT achieved superior performance in mRNA property prediction compared to previous methods, including on a novel influenza vaccine dataset.

Conclusions:

• CodonBERT represents a significant advancement in mRNA sequence design and optimization.
• The LLM provides a powerful tool for enhancing the development of mRNA-based vaccines and therapeutics.
• Further applications of CodonBERT can extend to predicting diverse mRNA-related biological properties.