AI for Bio · Model

AIDO Cell

Multi-scale, stateful virtual-cell world model, currently with two prototype cell types

Single-cell & perturbationAI-native biotechUnited StatesProduct only, no public artifact中文

Overview

NameAIDO Cell · aliases: AIDO Cell 1.0, AIDO.Cell, GB.Cell
Full nameThe name is a product name; the official page gives no expanded form
Ambiguous names (not auto-linked)AIDO
FamilyPart of the AIDO family
OrganizationGenBio AI
Organization typeAI-native biotech
CountryUnited States
Model classVirtual cell / multi-scale world model
DomainSingle-cell & perturbation (also: Genomics & RNA, Structure & co-folding)
First release
Basis: Official launch post with same-day press coverage
Note: One outlet gives 08-19 while a same-day report gives 08-18, which is used; the previous generation's single-cell weights appeared in 2024-11 and are a different product source
Other milestones
  • First public evidence: · Official launch post source
  • Official announcement: · Official launch post source
Latest update · no later event
Basis: No sourced substantive event since launch
Access is still a waiting list with the early-access programme in preparation and the full white paper available on request; a 2026-07-25 change to the previous generation's repository belongs to a different product line source
The most recent sourced event, not a claim that nothing changed since.
StatusProduct only, no public artifact · No public weights or reproducible materials, waiting list only
OpennessCode: Closed
Weights: Closed
This version releases no weights and its white paper is on request; the previous generation's single-cell weights are on the model hub under the developer's own community licence
Benchmarks reported0 counted (unconfirmed 1)
Wet-lab validationNone recorded
Unverified fields† training_data.a_public, training_data.d_wetlab_validation_only, background.funding
As of2026-11-16

What it does

Simulates DNA, RNA, protein and whole-cell behaviour in one shared state where perturbations accumulate and can be applied in sequence, currently for two prototype cell types only. Unlike transcriptome-perturbation models such as STATE and Lingshu-Cell, its comparators span structure, regulation and splicing models including AlphaFold 3 and AlphaGenome.

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. 0 of 1 pairings are counted.

BenchmarkRelationEval set / protocolSource typeNote and evidence
09 scIB (+Open Problems)Unconfirmed (not counted)—Unconfirmed sourceModel card claims zero-shot clustering SOTA; scIB not named, no numbers †disputed · evidence

Other evaluations (outside the 44)

  • Vendor's own virtual-cell evaluation (Public benchmark not among the 44) Self-reported: Designed, split and run by the developer across perturbation, knockout, structure, splicing and regulation, with self-chosen comparators and metrics; per-item values appear only in the on-request white paper
  • Replay of a known drug mechanism (Case study) Self-reported: Demonstrates a known kinase inhibitor's action in a leukaemia cell line; a case study, not prospective validation

Developers report: Developers report it is the only system covering all five task classes and matching or beating the strongest specialist baselines on most metrics of their own evaluation.

Independent evidence: No independent evidence found. The evaluation is vendor-built and vendor-run with no third-party reproduction, and neither weights nor white paper are public

Training data

Group in Figure 7: Not disclosed.

ClassValueDetail
a. Public databasesNot disclosedThis version's training data composition is undisclosed; the previous generation used public single-cell resources and is a different product source
b. Private or licensed dataNot disclosed source
c. Own wet-lab data used for trainingNot disclosed source
d. Wet lab used only for validationNot disclosedPublic material shows only a replay of a known drug mechanism, with no prospective wet-lab work mentioned source

Wet-lab validation

None. The developers report no wet-lab validation.

Background

  • Organization: The developer is an AI-native biology company headquartered in California with satellite labs in Europe and the Middle East, first presenting its digital-organism programme in 2024-12; founding year and headcount are not public.
  • Funding: Press releases say only that it is backed by leading investors; round sizes and lead investors are not public.
  • Products and partnerships: A first phase in 2024-12 covered six foundation models across DNA, RNA and single cells, and this virtual-cell product followed; researchers from several universities took part and no pharmaceutical partnerships are public.
  • Related models: STATE, Lingshu-Cell, scGPT / Geneformer

Reading notes

  1. A vendor-built, vendor-run evaluation is not a blind test: tasks, splits and comparators are the vendor's and per-item values are not public.
  2. A count aggregated across structure, splicing, regulation and perturbation cannot be converted into usability on any one task class.
  3. Only two prototype cell types exist and no prospective wet-lab validation has been seen.
  4. The previous-generation single-cell model and this version are different products and their figures must not be mixed.

Caveats

  1. Under this round's relation rules it enters the zero-pairing list: the model card's claim names no benchmark and gives no figures.
  2. Weights and full materials are not public, so no claim can be checked externally.
  3. Two dates circulate for the launch; this record takes the one from same-day coverage.

Sources

  1. Official product page (official page)
  2. Press coverage (launch day) (news) · 2026-08-18
  3. Previous-generation model card (model hub)
  4. Previous-generation preprint (preprint)

Corrections

1 correction
FieldWasNowBasisFound
benchmarks_reportedSelf-reported lead on this single-cell integration benchmark (self)Changed to unconfirmed: the model card does not name the benchmark, gives no numbers, and refers to the previous-generation productsource2026-11-16

Part of the AI for Bio benchmarks survey · all 112 models · open in the explorer. Record as of 2026-11-16.