The automated lab for agentic biology
Joined July 2021
- Tweets247
- Following1
- Followers5.5K
- Likes242
Pinned Tweet
How good is Claude at protein design?
Anthropic benchmarked their newest Claude models on protein engineering tasks, and we ran the wet lab work behind it.
We’re sponsoring wet-lab validation for the Defense Bio Hackathon, hosted by American Bioenhancement today in San Diego
180+ designers, 300 designs wet-lab tested in 5 tracks:
- Frontier bio
- Biological data
- Human performance
- Biomanufacturing
- Biosecurity
See you there! luma.com/kfgh2fvu
Challenge 3 of the Anthropic × Adaptyv Protein Design Competition is open. You have to design a protein that switches on GLP1R, the receptor that semaglutide (Ozempic, Wegovy) and tirzepatide act on in type 2 diabetes and obesity.
Entry is free and open to everyone. Designs must be de novo and 10–250 amino acids. We select with a Claude-based workflow, then make and test in our automated lab, and every measurement is published on Proteinbase.
Open until Sunday 25 October, 23:59 AoE. After feedback from the community we decided to make each Challenge 2 weeks instead of 1 week. So that means Challenge 2 (TNF-α) is still open until Sunday 18 October.
Submit your designs on Proteinbase: proteinbase.com/competitions…
All experimental results from @GoogleDeepMind's SynthID Bio model are now available on Proteinbase, check them out!
Google DeepMind just released SynthID Bio, a watermark written into the amino acid sequence of an AI-designed protein. We validated the proteins in our lab at @adaptyvbio and you can now find all the experimental results on Proteinbase.
In rheumatoid arthritis and Crohn's disease, immune cells in the joint or gut wall produce TNF-α continuously. TNF-α is active as a homotrimer, with three identical chains forming one complex.
Each of its three receptor-binding sites sits in the groove between two neighbouring chains. The site only exists in the assembled trimer, not on a single chain. One trimer then pulls three receptors together on the cell surface, and that clustering switches on inflammatory genes inside the cell.
They produce cytokines and draw immune cells drawn into the tissue which erodes the tissue. Adalimumab and infliximab block the receptor sites to stop it.
The second constraint: the binder also has to recognise mouse TNF-α, so the same design can be tested in mouse models.
Human and mouse TNF are about 79% identical, and adalimumab doesn't bind the mouse version at all. The epitope has to land on surface that's shared between the two species, at the receptor interface, with the pH-sensitive residues built into it - all on the same small patch.
Challenge 1 (EGFR) was about the inverse setup, meaning binding in acidic conditions, no detectable binding at 7.4.
Anyone can participate in this challenge on @proteinbase. Designs must be de novo, 10–250 amino acids. Up to 40 for Track 1, up to 20 for Tracks 2 and 3.
We select with a Claude-based workflow, then produce and test in our automated lab. Every measurement is published on Proteinbase.
Open until Sunday 11 October, 23:59 AoE: proteinbase.com/competitions…
Announcing Challenge 2 of the Anthropic x Adaptyv Protein Design Competition!
Humira, the best-selling drug ever, blocks TNF-α, an inflammation marker protein that’s overexpressed in rheumatoid arthritis and Crohn’s disease. In this challenge, the goal is to make a better binder for TNF-α that also releases selectively at pH 6.0 in the endosome of cells to improve drug half-life.
The protein must also bind both human and mouse TNF-α, making it easier to test the same design in preclinical models.
Two days left for Challenge 1 of the Protein Design Competition! You have until Tuesday, October 6 to submit your designs.
The goal of Challenge 1 is to design proteins that bind EGFR in the acidic environment of a tumour, but not at the neutral pH of healthy cells. Moreover, the proteins must bind to both the human and the mouse variant of EGFR to support easier preclinical testing.
Submit your designs on @proteinbase now: proteinbase.com/competitions…
We're launching Challenge 1 of the Anthropic x Adaptyv Protein Design Competition on @proteinbase: design a conditional binder to EGFR, a key cancer target.
The goal: bind EGFR under the acidic conditions of the tumor environment, but not at the normal pH around healthy cells. Moreover, the proteins must bind to both the human and the mouse variant of EGFR to support easier preclinical testing
Entry is free and open to everyone, with up to 20 designs per participant. We only accept de novo designs, meaning their amino acid sequences must be substantially different from those of known binders. Designs must be 10–250 amino acids long, in single-chain, scFv or Fab format.
We'll select designs using a Claude-based workflow, then make and test them in our automated lab. Every measurement will be published openly on Proteinbase.
Four more challenges will follow, with the next one launching next week!
Submissions close 4 October (end of day, AoE).
Submit here: proteinbase.com/competitions…
