DaltonTx launches AI antibody platform that folds millions of structures per hour

Overview

Updated Oct 2

Antibody discovery has long been trial and error: sequence millions of candidates, sort them by similarity, test the few that look plausible. On October 1, DaltonTx launched Dalton, which folds 87,000 antibody structures per hour—covering 2.6 million paired Observed Antibody Space (OAS) sequences in hours.

A chat interface runs Dalton and logs why each candidate was kept or discarded, so project knowledge survives team changes. Sygnature Discovery is checking whether Dalton would have reached candidate selection on a past oncology program with fewer compounds built. Bonito Biosciences uses it to design delivery ligands for central nervous system therapies.

Why it matters

If structural screening at this speed holds up, labs can prioritize antibody candidates by 3D shape instead of sequence similarity, cutting the experimental work needed to find a viable drug.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

87,000

Antibody structures folded per hour

Throughput achieved when Dalton folded the full 2.6 million paired OAS antibody space.

2.6 million

Paired antibody sequences folded

Entire Observed Antibody Space dataset folded in hours, not days or weeks.

Customers announced

Sygnature Discovery signed at the October launch; Bonito Biosciences announced its collaboration in May.

Voices

Curated perspectives — historical figures and your fellow readers.

Ever wondered what historical figures would say about today's headlines?

People Involved

Garry Pairaudeau

Co-founder and chief executive officer, DaltonTx

Leading DaltonTx through platform launch

Charlotte Deane

Co-founder and chief AI officer, DaltonTx

Leads AI model development; senior author of FlashABB

Organizations Involved

DaltonTx

AI drug discovery company

Launched Dalton platform October 2026

London-based startup building AI decision engines for drug discovery in both biologics and small molecules.

Sygnature Discovery

Contract research organization

Launch customer testing Dalton in oncology program

UK contract research organization integrating Dalton into its computational and medicinal chemistry toolkit.

Bonito Biosciences

Biotechnology company

Customer using Dalton to design delivery ligands for CNS therapies

Oligotherapeutic drug developer using Dalton to design precision delivery ligands for central nervous system targets

Timeline

2025 October 2026

3 events Latest: October 1st, 2026 · 1 week ago

October 2026
Dalton platform launches with antibody discovery capabilities
Latest Launch

DaltonTx launches integrated antibody design, structure prediction and validation workflow; folds 2.6 million paired OAS sequences at 87,000 structures per hour. Sygnature Discovery and Bonito Biosciences announced as customers.
October 1st, 2026
September 2026
FlashABB antibody structure model announced
Product

University of Oxford-developed FlashABB structure prediction model announced, generating hundreds of thousands of structures per hour.
September 9th, 2026
2025
DaltonTx emerges from stealth
Founding

Company launches with mission to build adaptive AI drug discovery platforms, led by Garry Pairaudeau and Charlotte Deane.
2025

Scenarios

Dalton's structural screening becomes standard practice in antibody discovery

Possible Resolves by Oct 1, 2027

Discussed by: DaltonTx leadership; Sygnature Discovery collaboration

If the Sygnature retrospective evaluation shows Dalton could have reached candidate selection with meaningfully fewer compounds synthesized, other CROs and pharma companies may adopt the platform. DaltonTx's position as a neutral tool provider—not a drug developer—would help it integrate into existing workflows rather than compete with them. The company's claim that data and models stay compartmentalized per program addresses a real industry concern about proprietary data use in AI environments.

Dalton remains a niche tool adopted by a handful of early customers

Possible Resolves by Oct 1, 2027

Discussed by: Pharmaphorum market analysis

The AI drug discovery field is crowded. Rivals like Recursion Pharma, Insilico Medicine and Schrödinger, plus tech giants such as NVIDIA and Alphabet's Isomorphic Labs, all offer competing toolkits. Dalton's differentiation rests on its decision-logging interface and structural-throughput claims. If early results are mixed or customers keep results private, the platform may not gain enough traction to move beyond its initial users.

FlashABB and structural convergence become an academic standard

Possible Resolves by Jan 1, 2028

Discussed by: University of Oxford researchers; structural biology community

FlashABB's open integration into Dalton could also push structural convergence—comparing candidates by 3D paratope shape rather than sequence identity—into wider academic use. Charlotte Deane's lab remains connected to Oxford and continues evaluating emerging antibody modeling approaches. If the method's results get published in peer-reviewed venues and adopted by other groups, Dalton could benefit from academic validation even if commercial adoption lags.

Shared industry developability models squeeze Dalton's differentiation

Possible Resolves by Q1 2027

Discussed by: Ginkgo Datapoints, Apheris, AbbVie, argenx, Lundbeck, Takeda via Drug Discovery News

On September 29, Ginkgo Datapoints and Apheris announced the Antibody Developability Consortium, with AbbVie, argenx, Lundbeck and Takeda as founding members. Members contribute antibody sequences; Ginkgo runs high-throughput wet-lab characterization and trains a foundation developability model, with the first dataset expected by early 2027. If big pharma comes to trust a pooled, consortium-trained model for spotting risky antibodies early, Dalton's developability screening competes with a shared industry resource.

Historical Context

2 moments from history that rhyme with this story — and how they unfolded.

November-December 2020

AlphaFold2 (2020)

DeepMind's AlphaFold2 won the CASP14 protein structure prediction competition, accurately predicting protein 3D structures from amino acid sequences for the first time at scale. The system's accuracy was a leap over prior methods, which had plateaued for decades.

Then

The protein structure prediction field was transformed; DeepMind released the AlphaFold Protein Structure Database covering nearly all known proteins.

Now

Structure prediction became a routine computational step in biology, and a generation of faster, more specialized models followed, including antibody-specific ones like FlashABB.

Why this matters now

Dalton's speed advantage—87,000 structures per hour for antibodies specifically—extends the AlphaFold trajectory from predicting single proteins to screening entire repertoires in hours, making shape-based selection practical for the first time.

January 2020

Exscientia's AI-designed drug enters trials (2020)

Exscientia and Sumitomo Dainippon Pharma announced the first AI-designed drug candidate to enter human clinical trials, an obsessive-compulsive disorder treatment called DSP-1181. Exscientia's systems designed the molecule after roughly 12 months of work, a fraction of the typical 4-5 years.

Then

The announcement validated that AI could meaningfully accelerate the design phase of drug discovery, attracting investment and attention to the field.

Now

Exscientia later merged with Recursion Pharmaceuticals in 2024. Garry Pairaudeau, DaltonTx's co-founder and CEO, was Exscientia's CTO before the merger.

Why this matters now

DaltonTx is the next step in the same thesis Pairaudeau pursued at Exscientia: instead of just generating candidates faster, Dalton tries to make the decision process itself faster and more transparent.