Deep learning-based codon optimization with large-scale synonymous variant datasets enables generalized tunable protein expression

Increasing recombinant protein expression is of broad interest in industrial biotechnology, synthetic biology, and basic research. Codon optimization is an important step in heterologous gene expression that can have dramatic effects on protein expression level. Several codon optimization strategies have been developed to enhance expression, but these are largely based on bulk usage of highly frequent codons in the host genome, and can produce unreliable results. Here, we develop deep contextual language models that learn the codon usage rules from natural protein coding sequences across members of the Enterobacterales order. We then fine-tune these models with over 150,000 functional expression measurements of synonymous coding sequences from three proteins to predict expression in E. coli. We find that our models recapitulate natural context-specific patterns of codon usage and can accurately predict expression levels across synonymous sequences. Finally, we show that expression predictions can generalize across proteins unseen during training, allowing for in silico design of gene sequences for optimal expression. Our approach provides a novel and reliable method for tuning gene expression with many potential applications in biotechnology and biomanufacturing.