Direct discovery · Biology · Chemistry
RFdiffusion-based workflow designs catalytic serine hydrolases
De novo serine hydrolase enzymes with experimentally measured catalytic activity and crystal structures matching design models.
Summary
A University of Washington-led team combined RFdiffusion with active-site and reaction-coordinate screening to design serine hydrolases from minimal active-site descriptions. Experimental assays and crystal structures showed catalytic turnover and close agreement between designed and observed structures.
AI role
RFdiffusion generated protein backbones for active-site motifs, while deep-learning structure tools screened reaction-coordinate compatibility before experimental testing.
Narrative role
This adds a 2025 protein-design follow-up showing AI-assisted design moving from static binders and folds toward enzymes with multistep catalytic mechanisms.
Caveat
The enzymes are model-system catalysts and still required expert workflow design, filtering, synthesis, screening, and structural validation.