DaltonTx opens an AI antibody discovery layer on its drug-discovery decision engine

London-based DaltonTx said it launched advanced antibody discovery capabilities inside the Dalton platform, tying AI design and structure tools to experimental feedback so teams can decide which antibodies to take forward.

Antibody discovery still burns a lot of wet-lab cycles on candidates that look good on paper. DaltonTx is selling a decision layer that keeps design, structure calls, and experimental feedback in one place so teams spend the next synthesis round on antibodies with a clearer reason to exist.

On Thursday, 1 October 2026, DaltonTx said it launched advanced antibody discovery capabilities inside the Dalton platform. The GlobeNewswire page is titled “DaltonTx, the decision engine for drug discovery, launches advanced AI-powered antibody discovery platform.” The page stamps October 01, 2026, 03:30 ET, which is 3:30 a.m. Eastern. The dateline is London. The source line is DaltonTx. An antibody is a protein the immune system uses to grab a target. Drug companies make antibody drugs to do that on purpose. A decision engine, in the company’s phrase, is software meant to help a team choose the next experiment, rather than only a model that draws a molecule. Those lines are the wire’s.

What the lines under the headline say the new capabilities are for. They say researchers can analyse, characterise, and prioritise antibody candidates at scale, and that integrated AI models and workflows support better informed decisions about which antibodies to progress experimentally. Analyse means study. Characterise means describe the properties that matter. Prioritise means put them in an order. At scale means across a large set, not one antibody at a time. Progress experimentally means take a candidate into the lab. The page uses those British spellings. Those lines are the release’s.

The workflow, as the release states it. The platform brings antibody design, structure prediction, and validation tools into a single workflow. Design is proposing the antibody. Structure prediction is a computer’s guess at its three-dimensional shape. Validation, here, is checking that design and that guess against what the lab actually measures. The release says the platform continuously tests AI-generated outputs against real experimental data, so researchers can pick candidates that are scientifically plausible and, in the company’s words, more likely to succeed in development. “More likely to succeed” is DaltonTx’s claim. The release does not print a success rate, a clinical result, or a regulator’s clearance. Those lines are the wire’s.

How a scientist uses it. Dalton is operated through a chat interface that captures the rationale behind each decision: why candidates were discarded, how priorities were set, and which goals drove each stage of the campaign. A purpose-built ontology links scientists’ observations and decisions to the specific antibodies designed and selected. An ontology, here, is a structured map of the terms and relationships in the project, so a note stays tied to the antibody it was about. The release says project knowledge then accumulates, rather than being lost between meetings, handovers, and team changes. Those lines are DaltonTx’s. The release does not print a transcript from a named campaign.

Dr Garry Pairaudeau, chief executive and co-founder, is quoted on the release. “Researchers today have access to an unprecedented range of AI models and computational tools for antibody discovery, but turning those outputs into confident scientific decisions remains a challenge,” he said. “We built these capabilities to bring data, models and expert judgement together in one coherent workflow. Ultimately, we're helping teams reduce complexity, accelerate discovery and focus resources on the most promising opportunities.” That quotation is his, in the release. The release does not print a price beside it, or a count of teams using the workflow.

The scale claim. The company said the platform can evaluate antibody repertoires comprising millions of sequences in a matter of hours. A repertoire is the set of antibody sequences under review. A sequence is the string of building blocks that defines one antibody. The company said it recently used the platform to fold the entire 2.6 million paired Observed Antibody Space at a rate of 87,000 structures per hour. Folding, here, means predicting the three-dimensional structure. Observed Antibody Space, often shortened to OAS, is a public collection of antibody sequences seen in real immune systems. Paired means the two chains of an antibody were recorded together. 2.6 million is the size of that paired set, as the company states it. 87,000 an hour is the company’s rate. Divide 2.6 million by 87,000 and that rate comes to about 30 hours. The 30 hours is the arithmetic on the company’s rate. The release states the rate. It does not state a clock time for the job, it does not say an outside lab timed the run, and it does not present the fold as a peer-reviewed benchmark. Those figures are DaltonTx’s.

Where in the work the release says the platform applies. Researchers can use it from hit identification and optimisation through candidate discovery, and for the design of complex antibody formats such as bispecifics. A hit is an early antibody that shows a useful effect. Optimisation, the release’s spelling, is the work of making that hit better. A bispecific is an antibody built to grab two targets. The release says the outputs are designed to inform scientific review and experimental planning, so teams can focus laboratory resources on the candidates with the strongest rationale for progression. A rationale, here, is the reason a candidate should go forward. Those lines are the company’s. The release does not name a drug that came out of the workflow, and it does not name a customer.

Professor Charlotte Deane, co-founder and chief AI officer, is quoted on the release. “Antibodies are the most successful class of medicines we have, yet we still discover them largely by trial and error,” she said. “There has never been a platform before that puts the necessary tools in the hands of every biotech, CRO and pharmaceutical company in the world. I co-founded DaltonTx to deliver the impact I knew this science could have, and I'm delighted to see that platform released today.” A CRO is a contract research organization, a company hired to run lab work. Biotech, here, is a company built around biological drugs. “The most successful class” and “never been a platform before” and “every” company are her words, in the release. The page does not print a ranking table under that quotation.

What the launch sits inside, as the release states it. DaltonTx calls the launch the latest step in a strategy to build drug-discovery workflows that connect advanced AI, experimental data, and human expertise. The company says it continues to collaborate with the University of Oxford and other academic researchers to evaluate emerging approaches in antibody modelling and analysis. The release names the University of Oxford. It does not name the other researchers, and it does not say Oxford certified the 87,000-an-hour figure.

Who the release and the company site say the co-founders are. Dr Garry Pairaudeau is chief executive and co-founder. The company site says he spent more than 25 years in drug discovery at AstraZeneca, including work that contributed to a marketed drug and a lead role in AI and automation, and that he was formerly chief technology officer at Exscientia. The site does not name that marketed drug. Professor Charlotte Deane is co-founder and chief AI officer. The site styles her Professor Charlotte Deane, MBE FRS. MBE means Member of the Order of the British Empire, a British honour. FRS means Fellow of the Royal Society. The site calls her an Oxford professor in BioAI, says she was formerly chief AI officer at Exscientia, and says she has more than 10,000 citations in BioAI. BioAI, here, is artificial intelligence applied to biology. A citation is another paper referring to her work. The 10,000 figure is the site’s. The release names the two roles. The years, the former titles, and the citation count are on the company site.

Who the about box says DaltonTx is. The box says the company is building the intelligence backbone for modern drug discovery, combining AI, human expertise, and experimental data into one continuous learning engine. It says the Dalton platform works across drug discovery for small molecules and for biologics. A small molecule is a drug made as a chemical. A biologic is a drug made from a protein or another product of a living system. An antibody drug is a biologic. The box says the platform captures what worked, what failed, and why, and that agents propose strategies, weigh trade-offs, and run a suite of AI, physics, and machine-learning tools. An agent, in that sentence, is software that can take a next step. The box says the company was founded on research from the University of Oxford and built by people with experience on AI discovery platforms at AstraZeneca and Exscientia. It says the product is secure, enterprise-quality engineering. That security line is the company’s. The about box does not attach an auditor’s report. Media inquiries go to Optimum Strategic Communications: Nick Bastin, Eleanor Cooper, and Henry Williams, +44 20 4604 4016, daltontx@optimumcomms.com. The company site lists an office at The Lighthouse, 368 Gray’s Inn Road, London WC1X 8BB, and a general address of info@daltontx.com. The release does not print a price, a funding round, or a customer list.

What the biologics section of the company site lists, kept as the site’s wording. It lists repertoire-scale ingestion of public and internal sequence sources, including the full Observed Antibody Space, annotated and clustered so patterns in a library can surface. Annotated means labeled. Clustered means grouped by similarity. It lists structure prediction, inverse folding, and protein language models that generate and optimise sequences, with fine-tuning from as few as 20 examples. Inverse folding means starting from a shape and proposing a sequence that would fold into it. A protein language model is a model trained on protein sequences. It lists binding prioritisation that ranks candidates on affinity, developability, and structural fit. Affinity is how tightly the antibody holds its target. Developability is whether it can be made and used as a drug. It also lists hit diversification: sequence-diverse antibodies that keep the same epitope binding mode, which the site says spreads immunogenicity risk and widens intellectual-property coverage. An epitope is the spot on the target the antibody grabs. Immunogenicity is the chance the patient’s immune system attacks the drug. Those lines are the product site’s description of the tools. They are not a measured result of the 1 October fold. The site also says Dalton runs in the customer’s own isolated tenant, behind the customer’s single sign-on, and that the customer’s data trains only that customer’s models. A tenant is that customer’s separate copy of the system. Single sign-on is the company’s existing login. Those sentences are the site’s.

The picture is the Structural Clustering screen on the DaltonTx site. A blue label at the upper left reads STRUCTURAL CLUSTERING. The canvas is white, speckled with faint gray points, a map of many predicted structures at once. The line along the bottom reads “A screening library mapped by predicted structure,” then a gold mark for known binders, a green mark for confirmed hits, and “click a cluster to zoom.” A screening library is the set of candidates a team is sorting. A known binder is one already shown to stick to the target. A confirmed hit is one the lab has backed. The frame does not print a calendar date. It is the product screen. It does not show the 2.6 million figure, and it does not show a price.

In plain terms, DaltonTx said on Thursday, from London, that antibody design, structure prediction, and validation now sit in one workflow on the Dalton platform, and that the workflow keeps testing the model’s output against lab data. A chat is supposed to record why a candidate was dropped, how priorities were set, and which goal drove each stage, with an ontology tying those notes to the antibodies. The company says the platform can score repertoires of millions of sequences in a matter of hours, and that it folded the 2.6 million paired Observed Antibody Space at 87,000 structures an hour. That rate and that count are the company’s. The release says teams can use the tools from an early hit through bispecific design, and that the output is meant to guide which antibodies go to the lab. The co-founders named on the release are Dr Garry Pairaudeau, the chief executive, and Professor Charlotte Deane, the chief AI officer. The release does not name a customer, a price, a financing round, or a regulatory clearance.