Adaptive Biotechnologies

The human team added 31 more.

Two outside labs, Adaptive Bio and Twist Bioscience, tested 211 designs.

17 bound their target, about 8%.

The two best, against a Mexican scorpion toxin called CN2, bound in the single-digit nanomolar range, near the site where a known neutralizing antibody sits.

At one point, the model autonomously emailed the protein sequences directly to the synthesis company Twist Bioscience.

Adaptive Bio and Twist Bioscience manufactured all 211 designs and measured whether each one actually bound to its target toxin.

17 met every criterion, binding success at both labs, every binding measurement at one micromolar or tighter, and no signal against an unrelated human protein used as an off-target check.

These are not hypothetical candidates sitting in a spreadsheet.

Working with Sutrine Research, the team guided a 64-hour design run.

Adaptive Bio and Twist Bioscience then tested 211 designs, with 17 successfully binding to individual toxins.

They haven't been shown to neutralize venom in cells, so this isn't an antivenom just yet.

Anthropic is releasing the designs and test data, though.

The application of this technology in generative protein design is fascinating, but sobering when subjected to real-world tests.

Adaptive Bio hosted a protein design challenge in 2025 to design a binder targeting the Nipah virus.

Of the designs tested in the challenge, 99 proteins were confirmed to bind, but none neutralized the targeted protein.

A specialized method called Mosaic, which combined multiple tools with expert tuning, outperformed general-purpose approaches.

Uh, went on to, you know, join Fred Hutchinson Cancer Center, becoming a, a faculty member at sort of the very early days of their computational bio work.

And then in two thousand and nine, co-founded Adaptive Biotechnologies, uh, where he served as the CSO for 16 years, building, uh, a really fascinating idea, you know, basically taking the enormous amount of information contained in the genetics of the adaptive immune system and applying that towards diagnosis and treating disease.

Uh, and now, uh, Dr. Robbins is embarking on a, a new chapter as president and co-founder of Herald Data, which is a company focused on probably one of the most enormous bottlenecks in the age of AI-driven science, and that is ultimately access to high-quality scientific data.

And Dr. Robbins is trying to create a, an entirely different model allowing these powerful AI and computational systems to use proprietary scientific data sets while ultimately allowing the organizations that generated them to retain control and participate in the value.

Yeah.

And, and, and those, um, you know, those were quite, a- again, early days in sort of the, the, the computational, uh, biology era.

You know, you're at Fred Hutchinson, but, you know, only a couple years later, and then you, obviously your, your timeline overlaps a lot in terms of creating adaptive biotechnologies.

But in itself, this part of the story, uh, you know, again, focusing on the adaptive immune system, which, you know, unlike the, the rest of our genome, you know, has infinite possibilities and, and changes a lot.