Lead use case
Drug testing
Reproducible human tissue and organoid models for toxicity and efficacy testing, aligned with FDA's shift toward New Approach Methodologies (NAMs).
The design engine for living tissue.
AI that predicts how to grow human tissue, and learns from every experiment.
Built on open science
- CZ CELLxGENE
- Human Cell Atlas
- Biohub ESM
- Geneformer
- scGPT
- Nicheformer
- BoTorch
An ESM for tissue recipes.
Protein models learned biology from sequences [1, 2]. For tissue, the sequence is a recipe.
Found by hand
A differentiation protocol can take months or years of manual experiments to optimize. [11]
A huge search space
Cells, scaffold, stiffness, growth factors, dose and timing all interact. One retinal protocol had about 200 million possible conditions. [11]
Scattered data
Outcomes sit in thousands of papers, recorded in incompatible ways. No general model of tissue formation exists yet. [10]
A tissue recipe, position by position
CCells
source, ratio, passage
SScaffold
chemistry, stiffness, pore size
MMechanics
strain, frequency, flow
GSignals
growth factors, dose
TTiming
order, duration, oxygen
Every experiment makes the next prediction better.
How the platform works- 01
Define the target
Specify the tissue, such as cartilage, liver or cardiac muscle, and the properties it must have.
- 02
Predict the recipe
Models propose ranked recipes (cells, scaffold, mechanics and a growth-factor schedule), each with a confidence score.
- 03
Test automatically
Recipes run on automated lab hardware. Outcomes are measured by imaging, single-cell and spatial sequencing, and functional assays.
- 04, then back to 01
Learn and improve
Results feed back into the model through Bayesian optimization, so each experiment makes the next prediction better.
Data foundation
Public atlases for breadth, and our own standardized experiment data for depth.
Foundation models
Open models that already understand proteins, cells and tissue neighborhoods, used as building blocks.
Recipe design engine
The model that turns a target tissue into ranked recipes and chooses which experiments to run next.
Closed-loop lab
Automated hardware that runs the recipes the engine proposes and measures what grows.
Where it applies first.
All applications
Regenerative medicine
Faster protocol development for cartilage, skin, bone and vessels, and eventually organ tissue.
Disease modeling
Grow diseased tissue, such as fibrotic tissue or a tumor microenvironment, to study mechanisms in human cells.
Longer-term
Personalized medicine
Patient-derived tissue, grown reliably enough to test which treatment works for that patient.
Why now.
Read the scienceRegulatory pull
FDA is phasing out mandatory animal testing.
The FDA Modernization Act 2.0 removed the requirement in 2022. FDA's 2025 roadmap names organoids, organ chips and AI models as alternatives, aiming to make animal studies the exception within three to five years. [17, 18]
Model pull
Open foundation models now exist.
For proteins (Biohub's ESM family), single cells (Geneformer, scGPT) and spatial data (Nicheformer). A tissue model can build on them. [4, 5, 6, 8]
Proof of concept
A robot found a better protocol in 143 runs.
RIKEN searched about 200 million culture conditions and improved retinal pigment epithelium production by 88%. A third-party result, on one cell type. [11]
Field roadmap
The field has a plan for virtual cells.
Leading researchers have set out how to build AI models of cells that span molecules, cells and tissues. [10]
Where we are.
Five stages, from a recipe language to a platform across tissues.
01 · We are here
Recipe language and data
A standard way to record cells, scaffold, mechanics, signals and timing, seeded from public atlases.
02 · Next
First recipe model
Trained on public data and scored against published protocols.
03 · Later
Closed-loop pilot
Recipes run on automated hardware with a partner lab.
04 · Later
First validated tissue model
A reproducible human tissue model for drug testing.
05 · Later
More tissues
Each new tissue starts from what the engine has already learned.
We are early stage, working on stage 01.
Questions we get.
What is working today?
We are early stage. We are building the recipe language and the data foundation first, then the first recipe model. We will publish results as we have them.
Which tissues come first?
We will choose them with our first partners. Human tissue models for drug testing, such as liver, are the lead use case.
How does a partnership start?
Tell us the tissue you need and what it has to do. We scope a first project together from there.
Will this count for regulators?
FDA's roadmap names organoids, organ chips and AI models as alternatives to animal testing. Whether a specific model is accepted depends on its validation, which is part of what we are building toward. [18]
Shape the first tissues the engine learns.
For pharma R&D teams, tissue engineering labs and investors who want to work with us from the start.