Antibody fragment leads optimised with Claude on affinity, stability and aggregation
Claude, connected to BigHat Biosciences' laboratory data system, proposed changes to two antibody-fragment drug leads and BigHat's automated laboratory made and tested about 1,010 variants in about 10 weeks; against the originals the best versions bound more than 10 times more tightly, the cancer-targeting lead stayed in solution at a temperature 18 degrees Celsius higher, and the lead meant to keep a drug in the body longer aggregated more than 100 times more slowly.
- Model
- Not disclosed; no model name or version given in the announcement
- Field
- Biology
- Date
- 2026-10-27
What was found
A drug lead has to bind its target tightly, survive manufacturing and storage, and not clump together. BigHat Biosciences pairs machine learning with an automated laboratory to improve such leads. Its scientists and an Anthropic researcher worked with Claude on two antibody fragments, one aimed at cancer cells and one intended to keep a drug in the body longer, and Claude, connected to BigHat's laboratory data system Reccy, proposed changes to each molecule. The laboratory made and tested about 1,010 variants in about 10 weeks. Partway through, the team came to suspect that the standard heat-stability and clumping metrics were misleading; Claude and the researcher reanalysed the raw data and proposed different metrics along with a protocol change to test them, and BigHat's scientists refined and ran that protocol, after which the announcement reports that the main predictions held. Against the original leads, the best versions bound more than 10 times more tightly, the cancer-targeting lead stayed in solution at a temperature 18 degrees Celsius higher, and the longer-lasting lead aggregated more than 100 times more slowly.
Novelty check
Searched Europe PMC on 2026-10-27. Pairing a language model with laboratory automation for antibody lead optimisation is an active area rather than a new one: "language model" AND antibody AND ("lead optimization" OR "affinity maturation") AND thermostability returned 49 hits, while the narrower antibody AND aggregation AND thermostability AND "machine learning" AND "automated laboratory" returned 2, a count that reflects the phrasing more than the size of the field. The novelty claimed here is accordingly not the method but the simultaneous improvement of three properties in two specific leads and, less usually, a model proposing replacement assay metrics and a protocol change that then held up. No existing registry entry covers antibody lead optimisation. No search was run against BigHat's own publication record and BigHat's technical report was not read for this entry, so this entry does not establish how far the campaign differs from BigHat's earlier ones.
Caveats and known objections
Entered at claimed under the rule that lab announcements start there. BigHat Biosciences ran the laboratory work with an Anthropic researcher embedded in the project, so there is no assessment by anyone outside the collaboration. Every improvement is stated against the original leads rather than against any state of the art, and no absolute values are given: no dissociation constants, no melting temperatures and no aggregation rates, only the ratios and the temperature difference quoted above. The announcement does not say how many of the roughly 1,010 variants Claude proposed as against the number from BigHat's existing pipeline, so the model's contribution to the measured gains cannot be separated from the automated laboratory's throughput. The statement that the main predictions held after the protocol change carries no numbers. The endpoints are surrogate properties of a molecule, binding, thermal stability and aggregation, and the two leads are described as antibody fragments that could become parts of future drugs rather than as drugs; there is no animal and no clinical evidence. BigHat's technical report is linked and was not read for this entry. Autonomy is graded ai-assisted on the strictest defensible reading: human scientists drove the project, set the three properties to improve, and refined and ran everything that was executed. The metrics episode is the strongest argument for collaborative, since Claude and the researcher reanalysed the data together and the model proposed the replacement metrics and the protocol change, and that reading is defensible; the weaker grade is taken because BigHat's scientists revised the proposal before running it and the announcement credits them with doing so. Field is biology rather than medicine because no therapeutic or clinical endpoint was reached.
No independent check is recorded on this entry yet.
Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Claimed and AI-assisted.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Community discussion
Graded claimed for verification and ai-assisted for autonomy. What these mean.
Cite this entry
whataifound.org. (2026). Antibody fragment leads optimised with Claude on affinity, stability and aggregation. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-10-27-bighat-antibody-fragment-optimisationBibTeX
@misc{whataifound-bighatbiosciences-2026-optimisation,
title = {Antibody fragment leads optimised with Claude on affinity, stability and aggregation},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by BigHat Biosciences / Anthropic. Verification: Claimed. Autonomy: AI-assisted.},
url = {https://whataifound.org/finding/2026-10-27-bighat-antibody-fragment-optimisation}
}