Enzymes are highly selective and sustainable catalysts for chemical synthesis, but their optimization is often limited by the difficulty of identifying functional starting points. This study shows that using the GenSLM protein language model to design TrpB variants can yield stable, active enzymes with broad substrate promiscuity, outperforming natural and evolved counterparts and demonstrating the potential of generative models to accelerate biocatalyst discovery.
- Théophile Lambert
- Amin Tavakoli
- Frances H. Arnold
