UW protein designers get more room to test new ideas

For protein researchers, extra computing power can mean something simple: trying an idea that would otherwise have to wait. A different training dataset, or a change to an AI model, might improve its designs. Running those experiments separately helps scientists understand which change made the difference.

On October 8, UW Medicine announced an in-kind grant of nearly seven million compute hours to its Institute for Protein Design. The Jen-Hsun and Lori Huang Foundation is providing access through CoreWeave using NVIDIA GPUs. Initial model releases are expected in late 2026 and early 2027, subject to progress and experimental validation. Goals include cancer treatments, durable vaccines and enzymes that break down persistent pollutants and plastics. UW announcement

What changes inside the research team

In a same-day interview with GeekWire, institute director David Baker described faster experimentation already underway. GeekWire reported that access began in June. Previously, computing constraints could push researchers to change several things at once, making it harder to isolate what helped. More capacity lets them explore alternatives more systematically. That account describes a change in how research gets done; it does not measure how quickly a medicine will reach patients. GeekWire’s interview

The scientific possibilities become clearer through the institute’s earlier work. Those studies provide context for this grant, rather than evidence of what the newly announced support has achieved.

What protein design can already do

In a paper published on December 3, 2025, researchers described RFdiffusion2, an AI method for designing the surrounding protein structure that holds an enzyme’s working parts in place. On a computer-based benchmark, it generated suitable structures for all 41 test cases, compared with 16 for the earlier method. The team also made and tested designed enzymes, finding candidates that carried out chemical reactions.

The distinction between those checks matters. A computer can identify a promising arrangement; a laboratory experiment establishes whether a molecule actually does something useful. The authors also reported an important limitation: their designed enzymes were less active than natural enzymes. More computing leaves that experimental challenge intact. Nature Methods paper

A separate paper, first published on May 21, 2026, explored proteins that attach to receptors involved in conditions including pain, cancer and migraine. Researchers designed small proteins that could activate or block these cellular switches. Structures of five receptor-bound designs closely matched their computer models. That gives researchers another way to investigate how a proposed molecule interacts with its target. Clinical usefulness would still need its own evidence. Nature paper

Making the tools useful beyond one lab

UW plans to share the new models, software and research data openly. Existing tools show how that can work: RFdiffusion2 already has a public code repository with installation guidance and example design tasks. Such resources give other researchers a starting point for trying the method, checking it and building on it. Availability alone does not supply the expertise or laboratory facilities needed to validate a design. RFdiffusion2 repository

This foundation grant has a different named funder from NVIDIA’s separate five-year, $1 billion science commitment, announced the same day. NVIDIA links its corporate commitment to the wider Genesis Mission. Neither announcement establishes that the UW grant is an allocation from that pledge; their values should remain separate.

For readers, the useful thing to watch is the chain of evidence: tools released, designs checked, experiments completed and results others can examine. Each step can make the next question easier to ask. The grant expands the opportunity to do that work.

Follow the evidence with Bright Weekly: five source-checked stories and one useful thing to try, free each week.

Related Bright coverage: NVIDIA’s five-year science commitment · The wider Genesis science commitments