Overview

An engine that designs the molecule that holds two proteins together.

Hold two proteins together and the cell gains a function it did not have. The design question is no longer what fits this site but what geometry brings these two into productive contact. The engine answers it: both proteins moving across their whole surface, every bridge enumerated rather than sampled, and only the designs an independent model can rebuild.

The molecule is the output. The engine, and the discipline that governs it, is the asset.

What the engine builds

A proximity drug does not act on two proteins. It acts inside a machine, target, effector, the scaffolding that holds them, and the designed molecule between. The engine designs into that whole assembly rather than into a two-body cartoon.

Modelled multi-protein induced-proximity assembly — **Illustrative.** A multi-protein induced-proximity assembly, rendered from our own structural models. The motion is an imposed path under elastic-network relaxation, restraints hold each protein’s fold while repulsion resolves overlap between them. It is not molecular dynamics and not evidence that this arrangement is populated. Molecular surfaces only; protein identities and the designed molecule’s chemistry are not shown.

Flexibility first

Backbone motion, side-chain repacking and anchor rotation run as one procedure across the entire landscape, never as a filter on a shortlist. A triage over a handful of arrangements is not a verdict about the rest.

Connectability as geometry

Whether two anchors can be joined is a topological question about the directions available to each, not about whether they happen to face one another. Searching only for facing pairs manufactures false negatives.

Design into the assembly

Bridge geometry and fusion chemistry are proposed against every component present, with the surrounding machinery treated as excluded volume. A bridge that is legal between two isolated proteins and illegal inside the complex is one you want to find now.

How we know the engine works

A design platform is judged on whether its verdicts hold, and a verdict is only as good as the sampling and the controls behind it. These are the checks, not the candidates.

It rebuilds known answers

Given only sequences and a compound, the ranking model reconstructs deposited induced-proximity complexes to within about two ångströms on the arm where it is licensed. Recovery on knowns is what earns an opinion about unknowns.

It searches to depth

Bridge feasibility is decided over the complete enumerated landscape, in one recent determination, more than 190,000 candidate arrangements, rather than over a ranked sample of it.

It returns clean negatives

When no legal bridge exists, the engine says so and says why. A decisive, controlled negative is a result the platform can act on, and a capability that separates a search engine from a wish.

It records what it has not shown

Where a method has not been validated for a case, that case is marked unlicensed and its output is not used. An insufficiently searched negative is recorded as not searched to depth, never as not there.

Five programs, one engine

Each engages more than one non-redundant disease driver with a single designed compound, rather than combining separate drugs. Nothing below is disclosed further here.

01 · 02

Dual-node degraders

Two programs, same design logic on orthogonal pairings. Landscape, ranking and provisional complete; compounds in synthesis at CRO partners. Assigned to TriaX Therapeutics.

Four-body assembly

Three proteins held by one molecule, on a purpose-built pipeline. Landscape and ranking complete across three composition classes; the composition application is drafted and files shortly.

Antigen-directed conjugate

The dual-target payload carried by an antibody to a tumour-selective antigen. Geometry and buildability gates passed.

Partner programme

Glue and conjugate pipelines designed on the engine against an external company's targets. Provisional filed; the composition stays here, the conjugate application is licensed to them.

Where they stand: what exists across all five is a ranked set of designed candidates with the provenance behind every judgement, not confirmed binders. Binding data is the gating milestone and nothing here has cleared it. Markers move when the work moves them.

See the portfolio →

The bridge is the design variable

The same two protein modules and the same pair of anchors, re-bridged at different lengths, produce materially different complexes. The bridge is not plumbing between two halves, it decides what the assembly becomes.

Cutaway of the assembly with an extended bridge between the two modules — EXTENDED The bridge holds the two modules apart. Proximity by tether.

Cutaway of the assembly with a compacted bridge, module surfaces approaching — COMPACTED A shorter bridge draws the surfaces together. Contact begins.

Cutaway of the assembly at the shortest bridge, drawn at the construction endpoint — FUSED The shortest bridge, at the construction endpoint. Whether two modules can be accommodated this closely is decided per pair.

Cutaway views of the same modelled assembly, under a single fixed camera so the three states are directly comparable. The designed compound is shown in black as a solid volume only, its chemistry is not depicted. Module colours denote the two halves of the assembly and carry no protein assignment.

The trade-off, stated: close contact needs modules sterically compatible with sitting against one another. Where they are too large or too awkward, the extended form is the correct design. In one recent determination the engine searched the complete landscape for a fusible arrangement and found none, a negative precise enough to act on. The axis is a range to choose along, not a ladder where tightest wins.

The specifics exist, they sit behind a CDA.

Target pairs, designed structures, ranked shortlists, scoring provenance and programme data are documented and available under confidentiality. Provisionals covering the platform and the partner-directed designs are filed; the one covering the four-body architecture and its pipeline is drafted and files shortly. If you evaluate these platforms for a living, the parts worth interrogating are sampling depth, gates and controls, and those we would rather discuss early than late.

How the engine works → · Talk to the team →