Pythia Labs launched on October 27 alongside a bioRxiv preprint describing Pythia-1, an all-atom generative model the company built to design antibodies from scratch.
The proof of concept targeted 22 therapeutic proteins, including G protein-coupled receptors, a family that has resisted antibody discovery for decades. After one computational design round, lab screening of about 48 candidates per target turned up binders in the single-digit nanomolar range.
Functional assays showed the designs could switch receptors on as well as off, an unusual outcome for generated antibodies. The characterised molecules also carried biophysical properties comparable to antibodies already in clinical testing.
What makes the model different
Pythia-1 reasons at once over amino acid sequence, three-dimensional atomic structure and explicit design specifications. That lets a team set where an antibody binds, which shape of the target it recognises, and development constraints, rather than generating candidates first and filtering afterwards.
The preprint is titled “Design of Full-Length Agonist and Antagonist Antibodies for Hard Targets”. The company published the full sequences of every lead antibody so outside groups can try to reproduce the results.
The caveats that matter
None of this is clinical data. The work covers binding and cell-based function in vitro, not efficacy in animals or patients, and the preprint has not been peer reviewed.
Designing an antibody that grips a target is only the first gate. Manufacturability, immunogenicity and dosing all decide whether a molecule reaches a trial, and the company has not disclosed a pipeline or a lead programme timeline.
