Questions about Proteus.
What we are building, and why. In plain terms.
What is Proteus?
Proteus is a biological intelligence: a system that designs biology rather than only describing it. Give it a target and it designs a molecule against that target, tests the design in its own lab, and learns from the result, from target to therapy.
What is a biological intelligence?
A biological intelligence learns the language and logic of living systems, from DNA and RNA to proteins and cells, in order to design biology rather than only describe it. Unlike a general-purpose model trained on human text, its native domain is life itself.
How is Proteus different from other AI drug-discovery companies?
Proteus runs the entire design loop, not a single prediction step and not a pipeline of services. It designs a candidate, tests it in a real lab, learns from the result, and designs again, until the molecule is real.
Is Proteus a contract research organisation or a software tool?
Neither. A CRO runs your experiment; a co-pilot assists your scientist. Proteus runs the design itself, from target to validated molecule, with the laboratory in the loop.
Where is Proteus based?
Proteus is based in San Francisco, where it runs its own lab. It designs de novo antibodies and proteins against a target and validates them in the loop, working with partners worldwide.
Can I bring a target to Proteus?
Yes. Proteus works as a de novo binder design partner: bring a target and it designs an antibody or protein against it, then validates the binder in its own lab by BLI. It is not a CRO or a service desk, the intelligence runs the design itself.
What is Proteus-SD?
Proteus-SD is a biological security foundation built to prevent pandemics, natural or engineered. It reads a new pathogen and designs the detection, the diagnostic, and the countermeasure, at the speed an outbreak moves.
What is Proteus-Discovery?
Proteus-Discovery generates new medicine. Given a target, it designs the therapeutic candidate and validates it in the lab, rather than screening for one by chance.
What is Proteus-N1?
Proteus-N1 is medicine designed for a single genome rather than a population, a therapy for one person. It is the horizon the rest of the work is built toward: n = 1.
What does lab-in-the-loop mean?
It means the laboratory is part of the model, not a step that happens afterward. Proteus designs, runs the experiment in a real lab, and learns from what comes back, so every cycle sharpens the next.
How does Proteus approach biosecurity and dual-use?
Proteus treats dual-use as a first-order constraint, not an afterthought. The same intelligence that designs medicine can defend against engineered threats, and Proteus-SD is built for exactly that: detection, diagnostics, and countermeasures.
What is de novo antibody design?
De novo antibody design means generating an antibody from scratch to bind a chosen target, rather than screening a library or immunising an animal. Proteus designs the binder computationally and then validates it in its own lab. In its first benchmark it produced validated binders for all four therapeutic targets.
What is de novo protein design?
De novo protein design is the creation of a new protein with a defined function, such as binding a target, directly from that target's structure rather than by modifying a natural protein. It is the foundation of how Proteus designs biology instead of only describing it.
Can AI design a protein or antibody from scratch?
Yes. Proteus designs proteins and antibodies against a target and confirms that they bind in the lab. Its first benchmark returned validated binders for all four targets, six of them under 50 nM by BLI. The same capability that designs medicine is why Proteus treats biosecurity as a first-order constraint.
How does AI design a binder against a target?
It starts from the target's structure, explores the molecular space around it, and generates candidate binders predicted to attach. A prediction becomes knowledge only when an experiment agrees with it, so Proteus tests its designs in the lab and redesigns from the result.
Is Proteus an autonomous or self-driving lab?
Yes. Proteus runs the design loop end to end, from design to build to test to learn, treating the laboratory as part of the model rather than a step that happens afterward. It is a self-driving lab for de novo antibody and protein design, with no scientist deciding the next experiment.
How is a biological intelligence different from AlphaFold or a protein language model?
Prediction models such as AlphaFold or a protein language model describe biology; they return a structure or a score. A biological intelligence designs it. Proteus is not itself such a model: it is the agent that runs the design, directing design tools against a target, judging the results, and validating them in the lab, closing the loop from target to validated molecule.
Is Proteus a protein design model like BoltzGen, RFdiffusion or Chai-2?
No. Proteus is the agent that runs the design, not a design model. It plans the strategy for a target, drives best-in-class design tools, scores the results, adapts mid-campaign, and learns from every campaign. The design models can be swapped in or out; the intelligence that directs them is the system.
How does Proteus compare to design models like Chai-2 or BoltzGen?
It is a different layer. A design model proposes structures; Proteus is the agent that decides the strategy, runs those tools, judges the results, adapts mid-campaign, and learns across campaigns. In its first benchmark this agent produced validated binders for all four therapeutic targets, six under 50 nM, then validated them in its own lab so each result feeds the next campaign.
Which targets has Proteus validated binders against?
In its first benchmark Proteus produced validated VHH binders against PD-L1, PDGFR, the insulin receptor, and TNF-α, all four targets, with six binders under 50 nM measured by BLI. These results predate any of them feeding back into the system.
What is an AI-native pharma company?
An AI-native pharma company is built so the system designs the therapeutic, rather than adding machine learning to a conventional discovery pipeline. The distinction is architectural: the AI decides what to make and test, and each experiment improves the next design. Proteus is AI-native, an agent that designs de novo antibodies and proteins by running design tools and validating them in its own lab.
What does "designed, not discovered" mean in drug development?
Conventional discovery searches libraries or natural antibody repertoires for a molecule that happens to bind a target. Designed, not discovered means generating a new molecule directly from the target instead of finding one that already exists. Proteus designs antibody and protein binders de novo, then confirms them in the lab.
How is AI-native pharma different from techbio and traditional pharma?
Traditional pharma is asset-centric, validating one hypothesis at a time through human-intensive lab work. Techbio adds data and engineering to speed that pipeline up, but still mostly screens for candidates that already exist. AI-native pharma makes design the core: the system designs the molecule, the lab tests it, and the result compounds back into it. Proteus works this way, as an agent that runs design tools rather than a model itself.
How are AI-designed antibodies validated, and what counts as a validated binder?
A validated binder is a designed molecule confirmed in the lab to bind its target, not only predicted to. Proteus validates by biolayer interferometry (BLI), which measures real binding affinity rather than a computational score. In its first benchmark it produced validated binders for all four targets, six of them under 50 nM.
Why design VHH (nanobody) binders de novo instead of screening for them?
VHH domains, or nanobodies, are single-domain antibodies: small, stable, and well suited to computational design. Designing them de novo means generating a binder directly from the target rather than immunising an animal or panning a library. Proteus designs in the VHH format, and tested ten designs per target in its first benchmark.
How can biology defend against AI-designed or engineered pathogens?
Defending against an engineered pathogen takes the same generative capability used to design biology, turned toward defense: a system that can read a new pathogen and design the means to detect, diagnose, and neutralise it. Proteus-SD is built as that layer, applying de novo protein and antibody design to threats that have not been seen before. An engineered pathogen need not resemble anything in nature, so defences that only match known agents are not enough.
Why can conventional screening and detection miss a novel engineered pathogen?
Most screening compares a sequence against catalogues of known threats, so something designed to resemble nothing on record can pass unflagged. Proteus-SD is built to reason about a pathogen from its structure and function rather than a database lookup, the same way a generative model reasons about the proteins it designs. For an engineered threat, the absence of a known reference is the whole problem.
How does Proteus-SD respond to a pathogen it has never seen?
Proteus-SD treats detection, diagnostics, and countermeasures as one loop rather than separate problems: reading a new pathogen, then designing the assay that flags it, the diagnostic that identifies it, and the antibody or protein that neutralises it. Because the candidates are designed computationally, the aim is to move at the speed an engineered outbreak would, rather than waiting on samples from infected patients.
How is a de novo design lab's approach to biosecurity different from monitoring and screening?
Biosurveillance and DNA-synthesis screening watch for threats; they do not design the response. Proteus-SD comes from de novo antibody and protein design, so its contribution is the countermeasure itself, the detection, the diagnostic, and the neutralising molecule, not only the alarm. Detection and design are complementary layers, and an engineered-pathogen response needs both.