Good news! The race is on!
"... Researchers expanded on a previous model largely built on amino acids, RFdiffusion2, by training it on a large swath of protein structures including backbones, side chains, and nucleic acids. The team even got as specific as the location of individual atoms or hydrogen bonds. The resulting model, RFD3, could design protein functional sites, RNA molecules, and enzymes from scratch, outperforming other models used to make protein components. In laboratory experiments, RFD3 designed two enzymes, one that broke down a toxic molecule and one that split another protein in two. ..."
From the abstract:
"The catalytic and other sophisticated functions of biomolecules arise from the placement of many atoms in three-dimensional space.
Here, we describe a generative artificial intelligence method for biopolymers, RFdiffusion3 (RFD3), which employs free atom diffusion to satisfy complex sets of atomic constraints more rapidly and accurately than previous methods across several challenges from enzyme design to RNA structure design.
We demonstrate experimentally the power of the RFD3 pipeline for enzyme design by designing de novo phosphotriesterases with a complex active site architecture incorporating a carbamylated lysine coordinating two metal ions and zinc proteases specific for the transthyretin protein involved in amyloid diseases; in both cases we obtain active catalysts in zero shot 96 construct design campaigns."
"Researchers at the UW Medicine Institute for Protein Design today [10/21/2026] introduced RFD4-Proteina, an open, generative model co-developed by IPD and NVIDIA Corp. The model uses biomolecular reasoning to improve how well scientists can design new proteins with specific functions. ..."
De novo Design of All-atom Biomolecular Interactions with RFdiffusion3 (preprint, open access)
Researchers [of the University of Washingto] achieve major advance in protein design (using Nvidia) "A new open-source generative model surpasses its predecessors by using biomolecular reasoning to solve difficult design problems."
Healthcare and Life Sciences Leaders Build on NVIDIA to Advance Drug Discovery and Expand Patient Care (Nvidia's original news release)
Fig. 1 All-atom protein design with RFdiffusion3.
a, Depiction of 100 step RFD3 diffusion trajectory, predicting both backbone and sidechain atoms. (top) noisy coordinates at different timesteps. (bottom) the network prediction at each time-step.
b, Generation of biomolecular interactions. (top) different conditions can be fed to the model. (bottom) sample output structures for each input condition.
c, Model architecture. Arrows show information flow, grey boxes indicate modules with learnable parameters. d, Inference times for RFdiffusion variants across length scales. RFD3 is an order of magnitude faster than previous versions of RFdiffusion because of its simplified architecture (runtimes are measured on NVIDIA A6000 GPUs).
No comments:
Post a Comment