// Compute in the loop

AI support for
biotech ventures.

Protein design used to be a search problem solved by hand and by luck. It is now a computational one. We apply AI across the design cycle, and we make that capability available to the ventures we back.

Applied to
Binders, antigens, constructs
Ranked for
Affinity, stability, expression
Loop
Design → build → test → design
Available to
Ventures in our programme

// The method

Next-generation proteins, designed with AI.

Sequence and structure models propose candidates; screening narrows them; laboratory results feed back into the next round. The loop is what matters — not any single model.

A design, build and test cycle

The useful unit is not a model, it is a cycle. A design round proposes candidates; in-silico screening removes the ones that will fail for predictable reasons; what remains goes to the bench; and the measurements that come back change how the next round is generated.

Run that loop well and each cycle starts from a better position than the last. Run it badly — or not at all — and you are testing candidates one at a time and hoping.

What we rank before anything is made

  • Affinity and specificity against the intended target and the likely off-targets.
  • Thermal and colloidal stability, and aggregation propensity.
  • Expression likelihood in the host you can actually use.
  • Developability — whether the molecule survives the process ahead of it.

What the models do not do

They do not replace the bench. A prediction is a way of deciding what to make next, not evidence that it works. Anyone who tells you otherwise is selling something. The value is in the ordering: fewer wasted syntheses, faster cycles, and laboratory effort spent on candidates that have already survived the cheap filters.

// The design cycle

Four moves,
repeated.

The same four steps run whether the output is a diagnostic binder, a therapeutic candidate or a vaccine immunogen.

  1. 01

    Computational design

    Structure-guided and sequence-based design of binders, antigens and engineered constructs.

  2. 02

    Screening & optimisation

    In-silico ranking for affinity, stability, expression and developability before anything reaches a bench.

  3. 03

    Design–build–test loops

    Laboratory results fed back into the models so each cycle starts better informed than the last.

  4. 04

    Venture-side AI support

    The same tooling and know-how extended to the ventures in our support programme.

// By area

What it changes,
area by area.

The method is shared. What it buys you differs depending on what you are building.

  • 01

    Diagnostics

    Binders with better specificity and fewer surprises between lots — designed against the panel the assay will actually meet. Diagnostics ↗

  • 02

    Drug development

    Candidate triage, so laboratory budget goes to the molecules that can survive diligence. Drug development ↗

  • 03

    Therapeutics

    Developability engineered in from the start rather than discovered during scale-up. Therapeutics ↗

  • 04

    Vaccine development

    Structure-guided antigen design, conformational stabilisation and epitope focusing. Vaccine development ↗