A foundation model for protein interaction and induced proximity. The engine carries both proteins in full flexibility across their whole surface, builds the molecule that holds them together, and keeps only the designs an independent model can rebuild.
Three axes, all flexibility, all spatial coverage, and induced proximity, held in one representation, so a single system reasons about protein interaction across targets it has never seen.
Hold two proteins together and the cell does something it was not doing before: degrade a protein, switch one off in place, relocate it, force a signal. That is a function added, not an activity subtracted, which is what makes induced proximity a modality rather than a workaround. The difficulty moves into geometry, and the name is literal: flexibility, spatial coverage and proximity, three axes of one problem, learned together rather than bolted on one at a time. Provisionals covering the platform and the partner-directed designs are filed; the one covering a four-body architecture and its pipeline is drafted and files shortly.
Most models pick one simplification, a rigid structure, a known pocket, a single target. TriaX refuses all three. Each axis feeds the same shared representation, so gains on one carry to the others.
Proteins are ensembles, not snapshots, so scoring runs against the full conformational landscape, transient and cryptic states included. Flexibility comes first, not as a correction afterwards: a search that samples first and flexes second only ever flexes what the rigid pass already liked.
Interfaces are broad and flat, and the site that matters often is not a known pocket. TriaX models the entire surface, so modulation sites surface anywhere on it.
A proximity drug acts inside a machine, not on a site: target, effector, the machinery around them, and the molecule between. TriaX designs into the whole assembly, which is also what makes a dual-target molecule possible at all.
The same engine proposes modulators and proximity-inducers for interactions it has never seen, instead of being rebuilt for every new program. That is the difference between a tool per target and a foundation.
How the engine works →
A model that cannot be wrong cannot be useful. Every prediction is checked against experimentally characterised complexes under the settings that produced it: a method that cannot recognise a known interface has not earned the right to rule one out.
Physical possibility is screened before compute is spent, every score is calibrated against reference complexes, and the limits are stated. That is what makes the positive results worth anything.
Read how we validate →
Two companies, cleanly separated. TriaX AI owns the engine and assigns each program outright to the company that develops it; only an outside party takes a licence. The engine stays reusable, and every program carries its own composition IP.
Geometry, flexibility and induced fit, the methods that turn two protein modules into a ranked set of designed molecules, and the corpus of evaluated arrangements those runs leave behind. Delivers ranked leads with the provenance behind every judgement, and keeps the platform.
Takes the assigned programs into synthesis and biology and owns composition of matter on the compounds. Discovery biology, DMPK and tox: the data that decides whether a designed molecule is a real one.
The platform company is not spending itself down on a single asset, and the company that develops a program owns it outright rather than renting a capability it cannot control. The engine also runs on other companies' targets as a service, where the designs belong to the partner and the methods stay here; the first such programme is under way.
See the portfolio →
The engine reasons about interaction geometry, not a target family, so the same process carries across areas and in both directions: degrade a protein, or hold a complex together.
Design is only half of it. Translational models and serious compute are what turn a ranked shortlist into a decision.
Patient-derived tumour organoids through a CRO partnership: primary-tumour and circulating-tumour-cell-derived lines for cell-based evaluation.
High-performance computing through the STAR industry program at TACC, behind the sampling and scoring the engine depends on.
Tell us the interaction, the modality and where the program is today, and we'll come back with next steps.