David Baker’s research team has successfully developed a cutting-edge generative AI model, RFdiffusion3, which leverages a unified all-atom generative architecture. This breakthrough overcomes the long-standing limitations in designing functional sites within complex biomolecules. When compared to previous methods, RFdiffusion3 significantly enhances both design precision and efficiency, expanding atomic-level design capabilities to encompass RNA and protein-nucleic acid hybrid systems.

The research team has experimentally validated two remarkable achievements. First, they engineered an artificial enzyme capable of hydrolyzing toxic organophosphorus compounds. This enzyme demonstrates a catalytic rate approximately 500,000 times higher than that observed under uncatalyzed conditions. Second, they designed an artificial protease with the ability to selectively cleave specific peptides associated with transthyretin, showcasing notable activity and selectivity toward the target substrate.

RFdiffusion3 opens up a new avenue for biomolecular design, with potential applications in environmental management, biomedicine, and industrial biocatalysis. However, further advancements are essential in areas such as catalytic efficiency, stability, and experimental validation to pave the way for practical applications.