Frontier AI for biological design
Today's generative models are exceptional at producing plausible structures. But plausibility is not enough when one residue can determine whether a molecule binds, folds, or fails.
ReasonBio puts reasoning inside the design process—so every generation begins with an understanding of the interactions that matter.
A dual-expert system turns molecular design from blind sampling into a deliberate, verifiable process.
Read sequence, structure, and context as one biological problem.
Identify the sparse residue-level anchors that govern function.
Co-design sequence and geometry under explicit biochemical constraints.
Test every candidate against independent models, structure, and physics.
Not bolted on as a ranker. Not added after thousands of candidates are generated. Our understanding expert makes explicit molecular commitments before—and during—generation.
Proteo-R1 is our reasoning-guided framework for de novo protein design. It separates molecular understanding from geometric generation, joined by explicit residue-level anchors.
Evaluated on the published RAbD benchmark under the same CDR-H3 redesign setting. Explore the paper for methodology, baselines, and limitations.
ReasonBio × Schekman Lab
Together with the Randy Schekman Lab at UC Berkeley, we are designing VHH nanobodies against the receptor-binding surface of Syncytin-2, a protein involved in placental cell fusion. The campaign turns a biological hypothesis into explicit designs that can be checked across independent models—and ultimately in the lab.
Blue: Syncytin-2 SU · Orange: designed VHH. Computational structure shown; binding and specificity require experimental validation.
A molecule either folds. An interface either forms. A design either survives contact with reality. Biology gives reasoning models something rare: answers that can be checked.
We are a team of reasoning-model researchers, multimodal foundation-model builders, and biological AI scientists working on one of intelligence's hardest frontiers.
The next reasoning frontier