LabGenius uses machine learning and functional assays to engineer antibodies

LabGenius Therapeutics combines machine learning with automated experiments to discover and optimise antibodies. Its EVA platform aims to balance several properties at once, including biological activity, selectivity and developability. The key evaluation question is whether the experiments represent the behaviour needed from the final molecule, rather than whether an algorithm can propose many possible designs.

01 / ProductEVA searches across antibody designs using experimental feedback

The technology page describes a loop that combines machine learning, automated laboratory work and multi-objective Bayesian optimisation. Antibody designs can incorporate different binding components, linkers and other elements. The platform uses measurements to guide subsequent proposals. This makes the experimental system central: what the assay rewards determines what the design loop is likely to pursue.

LabGenius emphasises testing functional activity and developability in addition to binding. That distinction matters for multispecific antibodies, where bringing components together can change behaviour. A design can bind the intended target yet fail to deliver the desired cellular effect or present practical development problems. The company’s stated approach is to evaluate combinations of properties rather than optimise a single convenient score.

The public description is a research platform, not a downloadable antibody design application. A buyer should distinguish platform capability from availability under a particular agreement. The appropriate deliverable might be an optimised molecule or a discovery programme, with supporting experimental data. It is not necessarily access to every model, automation component or proprietary assay used internally.

02 / AudiencePartners need a biological objective that can be measured repeatedly

A pharmaceutical team with a defined target combination and a meaningful functional assay is a plausible partner. It can specify the desired effect, unacceptable activity and relevant constraints. A useful programme might improve the balance between activity on intended cells and activity on cells that should be spared. The objective should be expressed in measurements that can distinguish a better design from a merely different one.

The partnerships page describes both discovery and optimisation routes. Discovery can start from a target, while optimisation can build on an existing molecule. Those routes should not be quoted as equivalent work. An existing candidate may provide a useful baseline, but a change to its structure can still require substantial re-evaluation of function and development properties.

The Absci blueprint provides another view of AI-supported biologic discovery. The Generate Biomedicines blueprint offers context on protein design programmes. LabGenius’s particular emphasis on experimental multi-objective optimisation is useful for comparison. Evaluate the assay strategy, molecule format and programme responsibility, rather than treating all AI antibody platforms as interchangeable.

03 / WorkflowProposed workflow: preserve the biological question through each design cycle

Start with a written description of the desired mechanism and the antibody format under consideration. This is a proposed evaluation workflow, not a Sequenced laboratory test. Identify the cell types, target expression ranges and controls that make the assay informative. A highly repeatable measurement can still be misleading if it represents the wrong biological context.

Agree the objectives and constraints before the first optimisation cycle. Activity, selectivity, stability and expression may pull the design in different directions. The team should know which trade-offs are acceptable and which properties are hard constraints. Otherwise an apparently improved aggregate score can hide a deterioration in the property that matters most to the programme.

Use the initial experiment set to establish a baseline and challenge the assay. Include known or reference designs where appropriate, and test whether results remain interpretable across batches. LabGenius describes experiments in the final antibody format. Confirm what that means for the proposed programme, including how components are expressed, purified and assessed together rather than inferred from isolated binding measurements.

After each cycle, inspect the competing designs and their measured trade-offs. A scientist should be able to explain why the next batch is useful. Some experiments should reduce uncertainty rather than simply pursue the current best-scoring design. Maintain traceability from the original construct to every functional and developability result used in a selection decision.

Before advancing a candidate, use evaluation conditions that were not simply the optimisation target. Determine whether the apparent advantage survives relevant variation in target expression, assay context or manufacturing conditions. The aim is to identify a robust molecule and a defensible explanation of its behaviour. A rapid design loop is useful only if it supports that broader judgement.

04 / PricingPartnership terms separate discovery work from later development rights

The reviewed pages provide no standard EVA software subscription or per-antibody rate card. The LG Chem agreement, announced in June 2026, gives a concrete example of the commercial model. It combines research collaboration with an option to license a resulting programme. An undisclosed upfront payment, possible milestones and royalties are described; it is not a public customer tariff.

The same announcement allocates stages between the partners and states that LG Chem funds research and development costs under the agreement. LabGenius’s work includes in-vitro preclinical assessment, while LG Chem undertakes further work including in-vivo studies before the option decision. That allocation belongs to this agreement. A new partner must negotiate its own scope, evidence package and development responsibilities.

The second Sanofi collaboration announced in December 2025 concerns NANOBODY optimisation for inflammation targets. It reinforces the relevance of the optimisation route without establishing a reusable price. A prospective partner should request terms for the actual target set and molecule format, including rights to improvements and obligations if a programme does not reach the intended milestone.

05 / DistinctionsThe distinguishing feature is optimisation across several measured properties

LabGenius is notable for linking a specific closed-loop antibody engineering approach with an owned pipeline and publicly named pharmaceutical partnerships. The platform’s multi-objective framing is useful because therapeutic design rarely has one sufficient score. A buyer can ask to see how the programme handles trade-offs, rather than accept a claim that every desired property improves simultaneously.

The pipeline page lists LGTX-101, a Nectin-4-directed T-cell engager, at an IND-enabling stage as consulted on 29 October 2026. LGTX-201 is shown at candidate selection. Those are development-stage descriptions, not clinical outcome reports. In-vitro selectivity or animal observations presented around the programmes should remain labelled by the setting in which they were obtained.

An owned pipeline also changes the partnering conversation. Ask which target spaces and rights are available, and how a proposed collaboration relates to internal programmes. The existence of company-owned assets does not itself create a conflict, but the commercial agreement should make relevant boundaries explicit.

06 / QuestionsAssay validity and downstream translation remain open questions

How well does the functional screen represent the disease context? Which performance properties can the platform measure directly, and which require later studies? What happens when a design improves one endpoint while worsening another? These are central scientific questions for an optimisation programme. They should be resolved with programme-specific evidence rather than broad platform throughput claims.

Clarify what data the partner receives, how results are reproduced and who owns newly developed assays or constructs. A useful handover must support the next development team. A molecule accompanied only by a headline performance claim is harder to assess than one with traceable methods, controls and the unsuccessful comparisons that explain its selection.

Finally, keep the regulatory and clinical sequence clear. IND-enabling work prepares evidence for a possible clinical development step; it is not an approved investigational application, successful trial or authorised treatment. The reviewed sources do not justify extrapolating platform claims into patient benefit. The commercial and research plan should preserve those distinctions.

07 / DecisionBegin with a measurable trade-off, then require evidence beyond the optimisation loop

LabGenius is worth examining when an antibody programme has several measurable objectives and a credible route into repeated functional testing. A good initial brief specifies the desired mechanism, available starting components and the evidence needed to choose a candidate. It also makes clear who will fund and perform the next stage after discovery.

Proceed when the assay, optimisation objectives and commercial rights fit the intended programme. If the biology cannot yet be measured reliably, improve that foundation before expecting the model to resolve it. If the immediate decision concerns clinical effectiveness, seek clinical evidence rather than treating a successful preclinical optimisation campaign as the answer.