Where we are aiming, and why
We are an early-stage company building a design engine. We do not have a clinical pipeline, and this page is not one. It sets out the target classes we are pursuing first and the evidence each would have to produce before we would call it a programme.
How a target gets on this list
Three filters, applied before we commit design capacity to anything.
A structural argument
There has to be a reason the binder should be small, synthetic and non-immunogenic rather than an antibody. Tissue penetration, reversibility, local delivery or conjugation chemistry.
A discriminating assay
If candidates cannot be told apart experimentally, the loop has nothing to learn from. An available, clean biochemical readout is a hard requirement.
An informative failure
We favour targets where a negative result teaches us something about the engine, not only about the target. Early on, that matters more than the size of the market.
Four target classes
For each, the argument for the modality and the evidence we would need before taking it further.
Secreted growth factors in ocular disease
Local delivery into a confined compartment is the friendliest possible setting for this modality. A molecule injected into the eye does not need to survive systemic circulation for long, the dose is measured in micrograms, and the target is a soluble protein in extracellular fluid rather than something behind a membrane.
Our interest is not in the indication itself but in what it tests. It is a setting where the design loop can be evaluated against a well-characterized soluble target with a straightforward binding assay, which makes it a good first proving ground for the engine rather than a good first drug.
What we would need to show
- A designed candidate that binds, with kinetics reproduced across independent surface preparations
- A specificity margin against related family members
- Stability in a vitreous-mimicking matrix over a defined window
- Activity in a cell-based functional assay, not binding alone
Coagulation proteases with a designed antidote
This is the prospect that uses the modality's distinguishing feature. An anticoagulant whose effect can be switched off within minutes by administering a complementary oligonucleotide is something a protein therapeutic cannot easily offer, and in surgical or bleeding settings that reversibility would be the clinical argument.
It also makes the design problem harder in a way we find useful. The binder and its antidote are a matched pair: the antidote has to out-compete the target for the aptamer's own structure. Designing both together is a genuine test of whether our structural reasoning is good enough to hold two objectives at once.
What we would need to show
- Inhibition in a plasma clotting assay
- Reversal kinetics with a matched antidote strand
- No activity against related proteases in the cascade
- Nuclease stability sufficient for a plausible infusion window
Cell-surface receptors for targeted delivery
Here the aptamer would not be the drug. It is the address label. Conjugated to a payload, a small binder that internalizes through a receptor enriched on tumour cells could carry cargo where a systemic agent should not go, and its size should help it penetrate tissue that an antibody conjugate reaches poorly.
The design objective is unusual, which is part of the appeal. Raw affinity matters less than internalization efficiency and the ratio of binding on target cells to binding on normal tissue expressing the same receptor at lower density. Conjugation site accessibility has to be a constraint from the start, so that attaching a payload does not destroy the binding face.
What we would need to show
- Receptor-dependent internalization in a cell assay
- Differential uptake against a low-expressing control line
- Binding retained after payload conjugation
- No measurable aggregation at working concentration
Structured RNA elements in viral genomes
Most oligonucleotide drugs that act on RNA work by base pairing and recruiting a degradation machine. This prospect is different: binding a folded RNA element structurally, recognizing a shape rather than a sequence, which should be harder for a virus to escape through silent mutation.
It is also where our folding models are most exposed, because both binder and target fold. That is deliberate. If the engine's structural reasoning is weak, this is where it will show first, and we would much rather find that out on our own work than on a partner's.
What we would need to show
- Binding to the folded element but not to the unfolded sequence
- Discrimination against the human RNA structures we screen
- An effect in a replicon or reporter assay
- A resistance profile across known sequence variants
What we would mean by each stage
Pipeline charts are easy to inflate, so here are the definitions we intend to hold ourselves to once there is something to put in one.
Every prospect on this page sits before the first of them. We will publish a pipeline when a programme has cleared hit identification, and not before.
Design
Target modeled, constraint specification approved, candidate libraries generated and scored. Nothing synthesized.
Hit identification
Shortlisted candidates synthesized and measured, with at least one confirmed binder whose kinetics reproduce across independent surface preparations.
Lead optimization
A binder family selected and iterated for off-rate, specificity, stability and manufacturability, with the modification strategy locked.
Enabling studies
Formal safety, manufacturing and regulatory work. A long way from where we are.
Bring us a target
Partner campaigns are how the engine gets tested against problems we did not choose ourselves, and the molecules stay with the partner.