LabGenius

EVA is the company. The machine learning engine explores the space of possible antibody designs, proposes candidates that balance several properties at once, and drives the automated build and test cycles that follow. The robotic lab exists to feed it. There is no separate product that would remain if the engine were removed, which is why the company describes itself as built around a machine learning driven evolution engine rather than as a lab with software attached.

EVA is described as an autonomous engine that runs design, build and test in automated cycles and learns from each round. The cycle is sketched, but the control around it is left blank: nothing states where a person reviews or overrides a proposal, what happens when an experiment contradicts the model, or which steps run without a human. The autonomy is a label rather than a described mechanism.

EVA is described as a machine learning driven evolution engine that uses deep learning neural networks to explore protein fitness landscapes, which conveys the idea without a named model, an architecture or a parameter count, and without characterizing the training data. A reader comes away with the concept and no detail on the system that implements it or when it last changed.

EVA is built in house alongside the company's robotics, with no outside model provider between a target and a design, so the chain is short. The pieces a partner would still ask for sit around it: the compute or hosting provider, any third party components, a subprocessor list, and the terms governing a partner's target data once it enters the system. With the model unnamed, none is stated.

What is public is a named partner, a named program and an operating cadence, rather than a result a reader can weigh. LG Chem signed a research collaboration and option in June 2026, funding early design work against a solid tumor antigen with milestones that could reach the triple digit millions, and the lead program LGTX-101 is named as a Nectin-4 by CD3 T cell engager built for tumor selectivity, the problem of hitting the tumor without harming healthy tissue.

LabGenius describes reaching a candidate after roughly four design and test cycles of about six weeks each, which is a concrete claim about how the loop runs. No clinical stage is stated for any program, no dated outcome figures are published, and there is no peer reviewed paper on how EVA performs.

For a partner, the live question is how its target and program data is retained, used in training or separated from other work, and LabGenius answers none of it. No protected health information is involved, since the inputs are protein designs and assay data.

LabGenius works with antibody and target data rather than patient records, so United States health privacy rules do not reach its work. The site states no business associate position and publishes no privacy document covering the EVA platform, so a partner weighing data handling starts from silence on this point.

LabGenius runs partner programs that put another company's target data inside its systems, so a security review would open by asking for an outside attestation. There is none: no SOC 2 report, ISO 27001 certificate or HITRUST certification, no trust portal and no security page.

The platform makes no device claim and is scoped to match. LabGenius designs drug candidates, so the regulatory position belongs to the molecules rather than to EVA. The lead program is preclinical, no clearance is claimed or needed for the engine, and any investigational new drug filing that follows sits with LabGenius or its partner as the drug's sponsor.

EVA has been shown designing antibodies in demonstrations, but a demonstration is a capability on display, not an evaluation of how reliably the engine works. No benchmark, failure analysis or audit of model behavior appears, and no responsible AI policy sits alongside it.

There is no way to tell from the outside how often an EVA design works, since LabGenius reports no hit or success rate with a method and a denominator. Nor does anything stand behind a design that fails: no guarantee, indemnity or correction route, with risk handled inside private collaboration agreements.

EVA is an internal engine that produces antibody programs rather than software a buyer runs, so there is nothing to connect to: no clinical record surface, and no electronic lab notebook, laboratory information management system, research data platform or partner connector.

The engine runs for LabGenius rather than being deployed to anyone, so there is no tenancy or hosting model to evaluate. Location is implied rather than committed: the company and its robotic lab are in London, and EVA and that lab are described as one integrated system. Nothing published names where partner program data rests or how one collaboration's data is kept apart from another's.

LabGenius earns through partnerships rather than selling access to EVA, so there is no price to quote. The public figures come from the LG Chem agreement: an undisclosed upfront payment and potential milestones in the triple digit millions plus royalties. Those are contract terms, so the business model is visible while the cost of a program is not.

Coverage is named and focused, with a program behind part of it. LabGenius concentrates on multispecific antibodies for solid tumors, where the problem it names is hitting the tumor without harming healthy tissue, and the lead program LGTX-101 puts a specific target pair behind that, Nectin-4 and CD3. The LG Chem collaboration adds a further solid tumor antigen, though unnamed. What is not established is clinical reach, since no program is confirmed to have reached patients in a trial.