AI for Bio · Model

IsoDDE

Isomorphic Labs closed drug design engine extending AlphaFold 3 co-folding and affinity

Structure & co-foldingAI-native biotechUnited KingdomProduct only, no public artifact中文

Overview

NameIsoDDE · aliases: IsoDDE, Isomorphic Labs Drug Design Engine
Full nameIsomorphic Labs Drug Design Engine
Ambiguous names (not auto-linked)None
FamilyNot a family entry. A single internal engine; AlphaFold 3 is its predecessor with its own record
OrganizationIsomorphic Labs
Organization typeAI-native biotech
CountryUnited Kingdom
Model classCo-folding, affinity, cryptic pockets
DomainStructure & co-folding (also: Small molecules, Antibodies)
First release · product update
Basis: Official article and Zenodo technical report released the same day source
Other milestones
  • First public evidence: · Official article source
  • Preprint v1: · Zenodo technical report (not peer reviewed) source
  • Official announcement: · Official article source
Latest update · no later event
Basis: No technical update found after release
No later sourced event found (as of 2026-11-16); the 2026-05 Series B is a corporate event, not a new technical release source
The most recent sourced event, not a claim that nothing changed since.
StatusProduct only, no public artifact · Internal and partner use only; no public code or weights
OpennessCode: Closed
Weights: Closed
Closed internal engine; only the official article and technical report are public
Benchmarks reported2 counted (self 1, retrospective 1)
Wet-lab validationNone recorded
Unverified fields† training_data.a_public, training_data.b_private_licensed, training_data.c_own_wetlab_for_training, benchmarks_reported[0], background.products_partnerships
As of2026-11-17

What it does

An internal drug design engine extending AlphaFold3: co-folding on cases with low similarity to training data, small-molecule binding-affinity prediction and cryptic-pocket identification. Like the open Boltz-2 it covers both structure and affinity, but it is closed and cannot be reproduced externally; it is the main reference point for open AF3 reproductions.

Benchmarks reported

Every row is one pairing from the survey's relation file; a score belongs to the benchmark page's ranking, where it is shown with metric, eval set, source type and date. 2 of 2 pairings are counted.

BenchmarkRelationEval set / protocolSource typeNote and evidence
30 PLINDER / PINDER / Runs N' PosesDeveloper report, official protocol—Self-reportedTech report claims a large gain in the hardest Runs N' Poses bin; no numeric page †partly verified · evidence
01 CASPRetrospective dataset use—Third-partyCASP16 blind affinity data used post hoc, figure only †partly verified · evidence

Other evaluations (outside the 44)

  • Novel antibody-antigen interface set (Own task set) Self-reported: Own low-similarity antibody interface set compared with AF3 and Boltz-2; vendor-reported
  • Affinity comparisons (FEP+, OpenFE, ChEMBL time split) (Comparison against baselines only) Self-reported: Bar charts by protein class, claiming to beat ML and even physics methods; no summary value in the text, and a second-hand figure is not found in official material
  • Retrospective cryptic-pocket case (Case study) Self-reported: Retrospective reproduction of a published experimental finding (cereblon case), not new validation

Developers report: clearly better co-folding than AF3 on low-similarity cases, affinity prediction beating all deep-learning methods, and cryptic-pocket identification.

Independent evidence: No independent evidence found. Closed; no third-party reproduction or blind-test result found, and blind rounds such as CASP17 have not reported

Training data

Group in Figure 7: Not disclosed.

ClassValueDetail
a. Public databasesNot disclosedThe report only says the cut-off matches AF3 (PDB before 2021-09-30); no dataset list source
b. Private or licensed dataNot disclosed source
c. Own wet-lab data used for trainingNot disclosedThe report lists no internal experimental data source
d. Wet lab used only for validationNoOnly retrospective cases on published structures, no new wet lab source

Wet-lab validation

None. The developers report no wet-lab validation.

Background

  • Organization: Isomorphic Labs was spun out of Alphabet in 2021, is based in London and is a sister organisation of Google DeepMind, positioned as an AI drug design company; AlphaFold 3 was released jointly by the two.
  • Funding: A first external round of $600M (2025-03) and a $2.1B Series B (2026-05), both led by Thrive Capital with Alphabet and GV; valuation undisclosed (company announcements).
  • Products and partnerships: Research deals with Eli Lilly and Novartis from 2024-01 (about $82.5M upfront combined, per trade press); the company says new funds go to expanding IsoDDE and moving candidates to the clinic.
  • Related models: AlphaFold 3, Boltz-2

Reading notes

  1. All figures come from the vendor technical report; the official article has none, and there is no third-party reproduction.
  2. "Beats AF3" is a relative comparison on the hardest similarity bin, not overall Runs N' Poses performance.
  3. Affinity is the commercially key claim, yet public evidence is comparison charts only, with no citable summary value.
  4. Being closed, outside checks must wait for partner disclosures or blind-test results.

Caveats

  1. Vendor-reported, non-peer-reviewed technical report; cannot be reproduced externally.
  2. The CASP16 affinity use is retrospective, not a competition entry.
  3. Affinity figures circulating second-hand are not in official material and should not be cited.
  4. Training data are undisclosed, so generalisation and leakage cannot be assessed externally.

Sources

  1. IsoDDE technical report (Zenodo) (preprint)
  2. Official launch article (company blog)
  3. Series B announcement (company blog)
  4. First external funding announcement (company blog)

Corrections

None recorded.

Part of the AI for Bio benchmarks survey · all 112 models · open in the explorer. Record as of 2026-11-17.