NVIDIA Model Designs Pesticide-Breaking Enzymes

Researchers at the University of Washington and NVIDIA used a new AI model to design enzymes that break down the pesticide paraoxon.

Updated on Nov. 18, 2026 in Biotech

NVIDIA Model Designs Pesticide-Breaking Enzymes

Baker's lab and NVIDIA developed the RFD4-Proteina model to design a pesticide-degrading enzyme. In laboratory tests, 89 of 288 RFD4-Proteina designs broke down paraoxon, while none of 192 RFD3 designs did.

Why it matters

The results suggest that expanding the number of amino acids at an enzyme's reaction site may help generate more active designs. The work points to computational protein design as a faster route to enzymes that target specific chemicals.

Laboratory assays found 89 of 288 RFD4-Proteina designs broke down paraoxon, compared with none of 192 RFD3 designs. The six most closely examined enzymes measured catalytic efficiencies 10 to 1,000 times those of earlier six-residue designs, and the extraction does not report field performance.

The players

NVIDIA

NVIDIA is a technology company best known for graphics processors that are widely used to train and run artificial intelligence models.

University of Washington

The University of Washington is a public research university in Seattle whose Institute for Protein Design has developed a range of computational protein-design tools.

The details

The RFD4-Proteina model generated multiple physically plausible arrangements of eight amino acids around a metal ion at the reaction site, compared with six in the earlier RFD3 designs. The computational design work took four days, and the researchers thought two additional amino acids might speed the reaction. Kieran Didi is the lead author of the paper, and Anthony Costa is NVIDIA's director of digital biology.

Timeline

  1. Researchers at the University of Washington used RFD3 to design paraoxon-degrading enzymes before the RFD4-Proteina work.

  2. Computational design of the eight-residue enzyme took four days.

  3. Didi worked with the team in Seattle for more than a year.

Big Picture

The RFD4-Proteina results extend the RFD3 model developed at the University of Washington Institute for Protein Design, which produced paraoxon-degrading enzymes with six reaction-site amino acids. The new work shows that adding two more residues produced active designs where the earlier RFD3 designs produced none.

Faster computational design of enzymes could eventually shorten the development time for tools that break down pesticides or other chemicals. The extraction does not describe any consumer product or availability timeline.

The takeaway

Expanding the number of amino acids at an enzyme's reaction site may be a useful design lever when building enzymes with computational tools. Researchers working on protein design can test wider active sites alongside existing six-residue designs.

Further reading

For related context, explore Biotech coverage on computational protein design.

Source note: This article includes information reported by Research & Development World.