The University of Washington’s Institute for Protein Design is on a hot streak. New studies, companies and spinouts are emerging at a rapid clip from the research hub, propelled by advances in artificial intelligence.
On Tuesday the latest venture based on IPD technology emerged: Xaira Therapeutics, a San-Francisco area startup co-founded by IPD head David Baker, backed with more than $1 billion from investors including Arch Venture Partners and Foresite Capital.
Only four days earlier, the IPD’s next-generation protein structure prediction and design tool — RoseTTAFold All-Atom — was featured on the cover of the journal Science.
And on Thursday, the IPD published a separate study in Science on designing specialized peptides, sleek ultra-small proteins that form the basis for Vilya, one of ten IPD spinouts.
AI-powered protein design is used to create new therapeutics, vaccines, biosensors, materials and more. The field is moving fast, and IPD research is behind many of the advances.
The pace of IPD research took off in 2021 with the release of RoseTTAFold, the institution’s open-access rival to DeepMind’s protein structure prediction tool AlphaFold. Both won Science Magazine’s Breakthrough of the Year award.
Since then, the IPD has released dozens of studies and a suite of open-access tools used to craft proteins of all sizes and shapes, from nanopores to target-seeking antibodies.
The IPD’s generative AI tools like RFdiffusion and RFantibody are being used to build Xaira, which has offices in Seattle and already has several employees in the city. The company aims to combine machine learning with large-scale data generation to fuel new models and develop new therapies.
Below is a list of other key IPD studies from the last ten months that are fostering new startups and creating a host of proteins with an array of shapes and capabilities.
• Introducing RFantibody, used to create designs for antibodies, powerful molecules that can recognize specific therapeutic targets. Released March 18 as a preprint, which has yet to be peer reviewed.
• Simplifying the construction of biomaterials with a new toolkit. Published March 13 in Nature.
• Creating hydrogels to bring proteins together in three-dimensional scaffolds inside or outside cells. Published Jan. 30 in the Proceedings of the National Academy of Sciences.
• Designing proteins that bind efficiently to biomarkers including human hormones, by deploying RFdiffusion and another IPD tool, ProteinMPNN. Published Dec. 18 in Nature.
• Generating carbon-rich minerals that have the potential to form the basis of carbon storge through engineered organisms. Published Dec. 14 in Nature Communications.
• Designing protein crystals, setting the stage for additional development that could lead to new optical tools, technlogies for chemical separation and other uses. Published Oct. 16 in Nature Materials.
• Designing protein fibers akin to those in silk, wool and spider webs, opening the door to new textiles and bioengineering applications. Published Sept. 4 in Nature Chemistry.
• Creating switch-like proteins that alter between two shapes and have the potential for use as environmental sensors or smart therapeutics. Published Aug. 17 in Science.
• Introducing RFdiffusion, which begins with an image of “pure static” that coalesces into an image of a protein. Published July 11 in Nature.
All of these studies were released within the last year and build on recent research from 2022 and 2023. This earlier research includes designs for custom enzymes, proteins that can slip through cell membranes, and a study showing how reinforcement learning, used in board games like chess, can support the design of new molecules. For more, see posts by the IPD and the Baker Lab.
