Antiverse

Generative design is the first and defining step of the platform. Models write antibody and nanobody sequences from the target sequence alone, with no known ligand, binder or earlier campaign required, and aim each library at the spot on the receptor that produces the intended effect. Engineered cell lines, functional assays and developability triage then test what the models propose, and those results retrain the models for the next round.

By the company's figure, its designed libraries reach a hit rate for functional agonists 2000 times higher than a naive library against one Class A receptor.

Every design passes through the wet lab before it counts. The loop has four steps: generative design, screening against the receptor in engineered cell lines, triage of binders by epitope and developability, and functional assays confirming agonist or antagonist activity. Results return to the models before the next cycle, which turns in about four weeks, and a partner receives a diverse, characterized antibody panel rather than raw predictions.

The decision rules for stopping a campaign or dropping a design set are not described.

The design engine pairs sequence based and structure based generative models with AI guided molecular dynamics. An analytical module then predicts structures for sequencing reads, maps the binding regions and clusters candidates for selection, and developability models screen for stability, safety and cross reactivity from the first round. Training draws on receptor data the company has gathered since 2017 and on its own screening results.

The models carry no names, versions or architecture details.

The generative and analytical models are the company's own, trained on its receptor data and screening results, and screening runs on cell lines it engineers in house, so partner targets and designs stay inside one company. Its scientists benchmark public structure prediction tools such as AlphaFold2 and Protenix in their research.

The compute and hosting behind the platform go unnamed.

Antiverse scientists led a 2026 paper in Briefings in Bioinformatics that tested nine structure prediction methods against 200 known antibody and antigen complexes and introduced AntiConf, a confidence score that raised precision and recall for every method tested.

Two conference posters report design campaigns with their counts. In one, more than 100 million in silico designs were narrowed to 77 antibodies, of which 14 bound the receptor, with affinities reaching 10 picomolar, and 9 were specific to the target domain. In the other, 103 of 520 binders against a difficult receptor were functional agonists. A 2025 mAbs paper on the PD-1 antibodies from the GlobalBio collaboration, with Antiverse's chief technology officer among its authors, reports two of them clearing colon tumors in mice and describes their selection from a synthetic phage display library.

None of this work tests the 2000 times hit rate figure itself.

Patient records do not enter the platform, which works on receptor sequences and structures, designed antibody sequences and the company's own screening data. Partner material is the live question: programs for Nxera, two large pharmaceutical companies, GlobalBio and the Cystic Fibrosis Foundation run through one closed loop whose results improve the shared models.

The privacy notice covers website visitors only, so neither the handling of partner data nor safety engineering for the models is described.

Antiverse is a United Kingdom company designing antibodies from receptor and sequence data, with no clinical program or patient data feed for the federal rule to reach. Its privacy notice covers visitors to the website under United Kingdom law.

The company takes no position on business associate status, or on any equivalent home regime, for the platform.

Partner targets, epitope designs and the company's receptor dataset all sit with Antiverse, across laboratory and office sites in Cardiff, Boston and Prague. Its legal pages cover the website and its terms of use.

No security certification, audit report or trust center covers the platform or the laboratory.

The platform is a discovery service run in house rather than a product used in care, so no device rules apply to it. Regulatory work begins with the antibodies. The PD-1 candidates from the GlobalBio collaboration are in nonhuman primate toxicology studies meant to support a clinical program, and the company says its antibodies are in investigational new drug enabling work. For licensed assets such as Nxera's, the partner holds development rights and would carry any filing.

No program has yet reached an investigational new drug filing.

Model testing shows up in the company's research. The AntiConf paper measured how accurately nine structure prediction methods reproduced 200 known antibody and antigen complexes and built a score to flag unreliable predictions of a binding pose.

How Antiverse's own design models perform from one receptor family to another, given training data that is denser for some families than others, has no reported results.

Design campaigns report their hit rates with counts. For one large pharmaceutical partner, 14 of 77 designed antibodies bound the receptor and 9 of those were specific to the target domain; in another campaign, 103 of 520 binders against a difficult receptor were functional agonists. Each design is built and tested before a partner sees it, so what reaches the partner is a characterized antibody rather than a prediction.

No guarantee or remedy tied to a failed design is offered.

Partners work with Antiverse through research collaborations rather than system connections. In the Nxera deal, Antiverse's libraries pair with Nxera's NxWave platform for receptor selection and structure work, and the Cystic Fibrosis Foundation takes promising candidates into its own laboratory to test in cells that naturally carry CFTR.

The platform has no link to clinical records or to a partner's notebooks, laboratory systems or data platforms.

Design and screening run in house, with the computational work and the wet laboratory joined in one loop, and partners receive antibodies and data rather than software to install. The registered office and laboratory sit in the sbarc spark innovation building on Maindy Road in Cardiff, with further offices in Boston and Prague.

Where partner data is hosted, and whether it moves between the British, American and Czech sites, is not said.

Revenue comes from research and licensing partnerships rather than software sales. Under the Nxera agreement, Antiverse receives an upfront payment and research funding and can earn milestones, while Nxera holds an exclusive worldwide license to the resulting antibodies with full rights to develop and sell them. The GlobalBio and Cystic Fibrosis Foundation collaborations follow the same partnered model.

Equity is the one figure on record: a $9.3 million Series A led by Soulmates Ventures, joined by Innovation Investment Capital, Domino Ventures, DBW, Kadmos Capital and the i&i Biotech Fund. No partnership carries a disclosed dollar amount.

Work centers on G protein coupled receptors, ion channels and other membrane proteins, a class of more than 400 disease linked targets that antibody drugs have largely left alone. Designs come in any format built on a heavy chain, from full length antibodies to single domain nanobodies.

The pipeline lists a checkpoint agonist in oncology with GlobalBio, an oncology antagonist and a central nervous system antagonist for two top 20 pharmaceutical companies, an agonist for Nxera, and wholly owned programs in thrombosis, idiopathic pulmonary fibrosis and metabolic dysfunction associated steatohepatitis. The Cystic Fibrosis Foundation work targets the small part of CFTR that sits outside the cell. None of the seven programs has entered a clinical trial.