Project Summary/Abstract: Recombinant proteins have a wide range of applications, from pharmaceutical products, drug discovery, protein-based polymers for drug delivery, antibodies enzymes, and sustainable technologies such as textiles and vegan food production such as the impossible burger. Towards meeting these increased demands by replacing batch culture with continuous culture reduces overhead costs, batch-to-batch variation, and increases protein production. Further, towards maximizing recombinant protein, yield computational techniques for gene engineering, such as codon optimization, use synonymous codon changes to increase protein production. Although codon optimization increases protein production in specific systems, synonymous changes to a gene sequence can cause unexpected detrimental results to the protein, such as protein misfolding, decreased protein yield, and vector loss. Therefore, codon optimization may not provide an optimal strategy for increasing protein production in batch culture and may introduce risk when scaling to continuous culture. CFDRC has utilized state-of-the-art natural language processing techniques to learn how synonymous codons are used by a target organism and apply this learning to gene engineering. We demonstrated that our AI-based codon harmonization model could predict the E. Coli synonymous codon usage with 73% accuracy, significantly above prior reports. Using this AI-based approach to gene engineering will provide an optimal strategy for increasing protein production in batch culture and continuous culture in addition to de-risk scaling up to continuous culture.

Project Narrative: Recombinant proteins have a wide range of applications, from pharmaceutical products1, drug discovery, protein-based polymers for drug delivery, antibodies enzymes, and sustainable products such as textiles2 and foods like the impossible burger. Recently, researchers have been critical of the gold standard approach to gene engineering, codon optimization, as the optimization process can affect protein conformation and function, and reduce efficacy and yield. CFDRC has utilized state-of-the-art natural language processing techniques to learn how synonymous codons and apply this learning towards gene engineering to provide an optimal strategy for increasing protein production and efficacy of therapeutics.