Intelligence, evolved

Evolution is the best search algorithm biology has. We are building the model that drives it.

Giving the search
a therapeutic direction

The question we’re asking isn’t how AI models can make the perfect drug.It’s how we can design the perfect evolutionary search.

Each target calls for a different balance of properties. Biophysics and chemical interactions guide how we shape the search.

Three molecular strands with a compact gold ring, a fused coral double ring and a branched green group at the same sequence position.

Create variation

Sequence changes create new combinations of molecular shape and chemistry.

A folded peptide binder fits into a complementary pocket in a textured blue protein surface.

Apply selection pressures

A useful candidate needs binding strength, specificity and stability. Selection brings these properties together around the target.

A peptide occupies a protein binding site while another molecular partner remains outside the occupied pocket.

Learn from biological effects

Laboratory results reveal where predictions succeed and where the search needs to adapt.

Smart directed evolution

We start with a diverse population of drug candidates and introduce mutations to create new variations. We assess their binding strength and specificity, then select promising candidates to seed the next generation. Each round explores new combinations of the properties a drug needs.

Our AI models learn from measured binding data to understand how a candidate’s shape and chemistry affect its interaction with a target. This learning helps guide selection by identifying which traits to favor, which to change, and how to balance them for each target.

Define a binding site and the molecular features the search must account for.

Generation 01

Our aim is to carry this learning to new targets, including those with little existing binder data.

Designing for scale

Scaling drug discovery means making reliable predictions for targets with very different shapes, chemistry and biological roles. Our goal is a shared AI model that learns across programs and adapts the evolutionary search to each target. That is how we aim to extend discovery across the human proteome.