Translating chemical reactions into enzyme sequences enables formate dehydrogenase engineering

REXzyme generated 20 soluble formate dehydrogenase variants that all showed CO2-reduction activity, with the top design reaching 2.5-fold higher specific activity than the reference enzyme.

• Why it matters: Enzyme engineering is expensive and often low-yield, and even with generative AI it is hard to bias sequence generation toward enzymes that perform a user-specified chemical reaction.
• What they did: The authors cast enzyme design as translation between chemical reactions and amino-acid sequences, then evaluated REXzyme with in silico metrics and tested it by conditioning generation on CO2 reduction.
• The result: In the formate dehydrogenase reverse reaction setting, reaction-conditioned designs were expressed and purified, and improved activity was achieved while all substitutions were distal to the catalytic site.

AI summary of the abstract · may contain errors · not medical advice · read the paper.