From the Research Universities edition of October 17, 2026
AI-designed proteases: 13 of 69 cysteine and 36% of 135 zinc protease designs cut their targets in laboratory tests
Phys.org (Science X) reporting on University of Washington (Baker lab) Nature papers · Higher education life-science research · United States
- Publisher
- Phys.org
- Original publication
- 16 October 2026
- Source retrieved
- 2026-10-18
What happened
David Baker's group combined RFdiffusion2 (RFD2-MI), ProteinMPNN/EnhancedMPNN and AlphaFold 3 to design proteases from scratch, then tested them in bacteria. Of 69 cysteine protease designs, 13 cut their targets; 36% of 135 zinc protease designs did. AI-guided refinement made the top zinc protease cut 30 times faster, and some designs cut disease-relevant targets such as TDP-43. The enzymes have not been tested in human clinical trials.
Why it matters
Shows a university lab pipeline where compute-heavy AI design is judged by experimental hit rates. It is relevant to research computing capacity planning and to how research offices describe validated versus computational claims.
Evidence and measured results
Secondary news report (Phys.org, 16 Oct 2026, edited and reviewed) of two Nature papers (DOIs 10.1038/s41586-026-11143-w and 10.1038/s41586-026-11167-2). Opened the report only; the underlying Nature papers were not opened in this run.
Limitations and uncertainty
Secondary reporting; figures were not checked against the Nature papers. Hit rates are for bench cleavage of designed targets in bacteria-produced constructs, not therapeutic effect. Lab-specific, with unreported compute cost.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-18; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Life-science groups need AI design compute tied to measurable laboratory validation, and administrators need to describe results accurately.
- Stakeholders
- Research computing director
- Core facility managers
- Principal investigators in biochemistry and structural biology
- Office of research communications
- Discovery questions
- How are GPU hours currently allocated to design workloads?
- How do computational candidates reach assay teams and results return?
- How are experimental hit rates reported externally?
- Value hypothesis
- Predictable GPU access and a defined compute-to-assay loop could shorten design-test cycles; the report does not measure cycle time or cost.
- Potential engagement
- A capacity and workflow assessment for one structural-biology or protein-design group.
- Unsupported claims
- Clinical or therapeutic benefit
- Time or cost savings
- That hit rates generalise to other enzyme classes or labs
Pre-sales engineering
Role takeaway
- Fit
- Fits GPU-heavy protein design groups with an assay partner; not a generic enterprise workload.
- Architecture and integration
- Chain design models, structure prediction and ranking on GPU queues; export candidate sets to assay teams and ingest results for the next round.
- Prerequisites
- GPU capacity and scheduler policies
- Licences and terms for the named models
- Candidate and assay data tracking
- Constraints
- Compute and software details not reported
- Secondary source
- Biosecurity review expectations not stated
- Security
- Report does not discuss security; add dual-use and data-access review for biological design work.
- Proof of value
- Replicate a small design-test round and compare hit rate to the reported 13 of 69
- Measure GPU hours per candidate that passes the assay
- Check that candidate lists and results are traceable
Delivery
Role takeaway
- Work
- Set up a recurring compute-to-assay workflow with shared tracking between research computing and a wet lab.
- Dependencies
- GPU allocation
- Assay capacity
- Model licences
- Ownership
- Principal investigator with the research computing lead as operational owner of the pipeline.
- Skills and adoption
- Computational and bench staff need shared vocabulary for candidates, hit rates and failure modes.
- Governance checkpoints
- Biosafety and dual-use review before design campaigns
- Review of reported hit rates against raw assay data
- Provenance records for model versions
- Acceptance criteria
- Proposed: each round logs candidates, assay results and model versions
- Proposed: hit rate is reported per round
- Proposed: claims distinguish bench results from clinical relevance
- Risks
- Overstating results beyond bench assays
- Compute bottlenecks stall rounds
- Unclear ownership across compute and lab teams
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
The workflow chains several AI models, structure prediction checks and wet-lab assays, implying GPU capacity plus tight links between computation and experimental teams.
Governance
Who approves, reviews and stays accountable for outcomes?
Report experimental hit rates alongside computational claims, and keep clear that clinical relevance is not demonstrated.
Security and privacy
What data, permissions and controls need testing?
Not addressed by the report. Biological design tools raise dual-use questions that the report does not discuss.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Needs staff who span AI model operation and assay work; the report says nothing on training or workforce.
Procurement
What should contracts, pricing and exit terms secure?
Capacity planning should include GPU queues and model licensing; the report gives no cost data.
Operating model
Which teams own the service once it runs?
Success depends on a repeated compute-to-assay loop with shared ownership, not on computation alone.
Publication history
- 2026-10-17Research Universities · Issue 396 resources
Stable resource ID: phys-org-ai-designed-proteases-hit-rates-202610