Claude enters the antibody design loop
-
Nº CXLI
- Date
- 12 Nov 2026
- Issue
- 141
- Stories
- Four
- Editor
- ARC
Four preprints today, and the throughline is agents taking over the parts of discovery people usually eyeball.
Claude helps design an anti-IgE antibody for food allergy
Claude helped design an antibody meant to disrupt IgE in food allergy, in a new bioRxiv preprint that calls the candidate best-in-class against existing IgE blockers. The design campaign ran on Claude, Anthropic's general-purpose language model, rather than a protein-specific design network trained only on sequences and structures. IgE blockade is an established clinical mechanism, so a model-assisted candidate here can be judged against drugs already given to patients instead of against a computational score. The preprint stops at the design and characterization stage, not clinical data.
ProTEUS scores single mutations to steer protein evolution
ProTEUS models protein evolution one mutation at a time, in an arXiv preprint describing what its authors call a world model — a system that predicts the consequence of an action before the action is taken. Here the action is a point mutation and the prediction is the resulting change in the protein. Proposing variants and scoring them happen in the same system, so rounds of directed evolution can be planned in silico before anything reaches the bench.
DE-Rigor-Agent audits assumptions in RNA-seq analysis
An agent checks assumptions before running differential expression on bulk RNA-seq, in a bioRxiv preprint. DE-Rigor-Agent tests whether a dataset actually meets the conditions the chosen statistical model requires, then adapts the analysis when it doesn't, instead of returning a gene list regardless. Differential expression is among the most-run analyses in biology and among the most frequently run on defaults, so putting the assumption checks inside the workflow changes what counts as a finished result.
ANTIOCH-VS screens from protein sequence to candidate hits
ANTIOCH-VS runs virtual screening end to end, taking a protein sequence and returning ranked hits through an open-source ensemble of structure and docking steps, per an arXiv preprint. It is pitched as modular: components can be swapped out without rebuilding the pipeline around them.
Reply with your discoveries. A human reads them. Forward freely.
|