Despite progress in computational protein design, creating functional de novo enzymes—especially those with complex, multistep reaction mechanisms—remains challenging. Success depends on precise active-site geometry and dynamic compatibility across the entire catalytic cycle. Lauko et al. introduce a strategy combining deep learning–based protein generation (RFdiffusion) and ensemble modeling (PLACER) to assess active-site preorganization throughout all reaction steps.
Using serine hydrolases as a model system, the researchers designed novel enzymes from scratch, achieving high structural accuracy and catalytic efficiency. By explicitly optimizing catalytic geometries rather than relying on directed evolution, they showed that including full catalytic triads and oxyanion holes is essential for turnover.
Their results demonstrate that tailor-made backbones and evaluation across all intermediates and transition states significantly improve enzyme design success. The approach is broadly applicable to other multistep enzymatic reactions, offering a new pathway to effective enzyme creation.
More reading;
Computational design of serine hydrolases
SHARE
