AI-designed enzyme advances sustainable synthesis of HIV drug

Scientists from the University of Washington’s Institute for Protein Design, in collaboration with NVIDIA and researchers at The University of Manchester’s Manchester Institute of Biotechnology, have used a new AI model to design enzymes capable of supporting the synthesis of a global health drug, providing a significant proof of concept for computational protein design in pharmaceutical manufacturing. The model, RFD4-Proteina, combines the Institute for Protein Design’s RosettaFold Diffusion line with NVIDIA’s Proteina-Complexa to generate entirely new protein sequences and structures in silico. It concurrently refines a protein’s amino-acid sequence, three-dimensional structure, and interactions with target molecules, allowing it to consider multiple aspects of structure and function at the same time. The system can also allocate more computational effort to promising designs, improving the efficiency with which researchers identify functional candidates. Across computational benchmarks, the model achieves design success rates five to ten times higher than previous approaches and generates protein candidates five to ten times faster. The Manchester team, led by professor Anthony Green with Dr. Henry Jones and Dr. Zefei Xu, tested the model on the synthesis of lenacapavir, a drug used for HIV prevention. The researchers used RFD4-Proteina to design de novo enzymes able to catalyze a key amide-bond-forming reaction. Of 67 designs selected for experimental validation, four showed clear enzymatic activity. The most effective design converted 79% of a bulky chiral amine into the desired amide product and demonstrated the required selectivity for the S-enantiomer. The reaction is considered challenging because the enzyme must favor the intended amide formation while suppressing a competing reaction with water, which is present in large excess. The work, published as a preprint, demonstrates that AI can design new enzymes for difficult steps in active pharmaceutical ingredient synthesis. The researchers said the results show how advanced generative models can accelerate the development of biocatalysts for medicines and other essential chemicals. The model is open source under a BSD-3 license, enabling broader use in protein science, medicine, sustainable chemistry, materials, and industrial biotechnology.