SynthyraOur vision
One model of proteins. Every service runs on it.
Atlas is one model that reads any protein. Biosecurity screening, host-pathogen prediction and the rest of our services all run on it, so every improvement to the model reaches every service. Each service is also an API that people and agents can call.
The thesis
Design for the outcome in the patient.
Binder design works. A target comes in, a molecule is designed to engage it, and the interface is scored. It is a well-posed problem with real progress behind it.
It is also a narrow one. How well a molecule binds is one part of how it behaves. The disease response, the toxicity and the side interactions all happen outside the binding site.
The canonical isoform of every human protein is known. Protein and ligand relationships can be predicted at proteome scale. Genome-scale models of human metabolism are published and open. If a disease outcome is the goal, that context belongs inside the design problem.
Most design assumes that hitting the target produces the outcome. We score the outcome directly.
The objective
What we score a design on.
A model of the whole system is only useful if you can ask it a question. We write the questions we care about as terms in the objective, so a candidate is scored on what it does to the whole system.
01
Efficacy
Whether the candidate moves the system toward the state you were trying to reach.
Abstracted from predicted interaction networks and pathway context.
02
Toxicity
Predicted liability resolved by tissue and compartment. Where a molecule is active matters as much as how much.
Abstracted from localization annotation, predicted target breadth, and liability proxies.
03
Interaction burden
The other interactions a candidate adds to a system that already has many, including interactions with other drugs. We score them up front.
Abstracted from protein-ligand screens and published genome-scale metabolic models.
Score the candidate on what it does to the system.
Company trajectory
Where the company goes next.
Atlas is in use today. Biosecurity screening and host-pathogen prediction are the first markets. Design programs follow, with partners, on the same models.
01
Operating now
Atlas platform and API
Protein interaction, ligand and annotation tools that researchers use today.
02
First markets
Biosecurity screening and host-pathogen prediction
Screening that finds sequences of concern current checks miss, for synthesis providers and government, and prediction of which host proteins a pathogen acts on. Both run on Atlas.
03
Next
Design programs
Designing proteins and small molecules with partners, scored on their effect on the whole system.
04
Long horizon
Novel asset development partner
Co-developing new therapeutics, enzymes and other biological assets with partners, with Synthyra holding a stake in what it helps create.
Beyond therapeutics
The same approach works for metabolic engineering.
The objective does not have to be a disease. Ask for more of a molecule a cell produces, keep the toxicity terms, and the same machinery designs for metabolic engineering.
Carbon fixation is the clearest case. Rubisco is the enzyme everyone names, and it is slow. But an organism is a network. Speed up Rubisco alone and the bottleneck moves to the next step.
We ask for the product, and let the model decide which enzymes to change.
01
Every enzyme in the cycle at once
Every protein in the Calvin-Benson-Bassham cycle is a design variable, and the score is what the pathway produces.
02
Named products
Sugars and sugar intermediates, glycolic acid, and precursors to fuels and materials are all concentrations a model can be asked to raise.
03
The same liability terms
Byproduct toxicity and host burden stay in the objective. A pathway that makes more of something and kills the organism has not been improved.
The applications include crop yield, nutrition and carbon capture.
The next step is in the lab.
We are looking for partners who can test designs experimentally, and for strategic partners.
Start a program conversation