From prediction to creation: How AI-driven design is reshaping microbial biocatalysis.

Abstract

Enzyme engineering has long relied on rational design and directed evolution, yet both remain bounded by natural enzyme space - a constraint felt acutely across the largely uncharacterised microbial world. The 2021 advent of accurate structure prediction marked an inflection point, catalysing AI methods that now extend far beyond folding. These tools are organism-agnostic; the microbial domain is their proving ground because it supplies both the uncharacterised sequence space and the expression hosts. Here we examine three stages: mining microbial sequence space; generative design, whether conditioned on structure or sequence; and machine learning-guided directed evolution. We trace the trajectory from reading the protein universe toward writing it, asking where each has delivered for microbial biocatalysis. Because published claims are hard to compare, we propose a tiered framework - catalysis, selectivity, producibility, process fit - for judging whether an AI-designed enzyme is a useful biocatalyst rather than merely a functional protein.