Skip to main content
model-family passport · Review date not recorded

scGPT

A GPT for the language of single cells.

4/7Evidence fields documented
60-SECOND EVALUATION VIEW

What should a scientist know before using scGPT?

SupportedEvidence supports the stated context with explicit boundaries
Best suited forSimulation · Prediction
Evidence supportsSingle-cell downstream tasks: Peer-reviewed
Evidence does not establishUniversal superiority, therapeutic success, clinical utility or regulatory acceptance.
Major limitationPerformance depends on the evaluation dataset and operating conditions.
Current registry recordVersion history not yet curated1 recorded release · Review date not recorded. A newer version is not assumed to be universally better.

What it is

scGPT is a generative pretrained transformer over 33M+ single cells that transfers to cell-type annotation, batch integration, perturbation prediction and gene-network inference. It helped popularize the 'foundation model' framing for single-cell genomics.

Evidence trail

BioAtlas keeps the path from source to decision visible. A connection records provenance; it does not imply that evidence is sufficient for every context.

Sources4 connectedPrimary resources and normalized claims
Claims1 normalizedSingle-cell representation learning
EntityscGPTmodel-family · Version history not yet curated
ReviewReview date not recordedReview date not claimed
ConclusionContext requiredAdd to an evaluation before operational use

Model passport

Entity typemodel-family
OrganizationUniversity of Toronto (Bo Wang Lab)
Model family introducedNot normalized
AccessOpen source
Commercial useAllowed / verify checkpoint terms
DeploymentSelf-hosted
ComputeGPU recommended
Domainscells
Biology → representation → computation → evidence

How scGPT represents biology

model-familycells

Category is navigation. These fields describe the model-specific computational transformation and deliberately override broad category defaults.

1 · Biological inputs
Single-cell or perturbation data
2 · Input representation
Gene-expression representation
3 · Internal representation
Cell-state embedding
4 · Architecture
Single-cell foundation model
5 · Learning objective
Cell representation / transition learning
6 · Output representation
Cell embeddingGene-expression state

Biological scale

Modalities & tasks

CellTranscriptomicsSimulationPredictionRepresentation

Registry, claims and frontier intelligence

Versioned registry

Version history not yet curated

1 version record · release year not yet normalized. Model-family identity remains separate from capability and access changes.

Explore version lineage →

Connected research frontiers

These records describe active research directions, not guaranteed capabilities of this model. Evidence stages and unresolved questions are preserved separately.

Virtual cells

Virtual cells that predict perturbation response

Arc Institute · Virtual Cell research community · 2026-04-30
Recent preprint

Can models forecast how cell populations respond to unseen drugs, gene edits, cytokines and environmental changes across biological contexts?

Evidence boundary and unresolved questions

Recent strict evaluations show marked performance drops under unseen contexts and metric-dependent rankings; simple baselines remain competitive on some global trends.

  • Can models recover perturbation-specific mechanisms rather than average expression shifts?
  • How should cell distributions, dose and time be represented?
  • Which metrics predict prospective experimental usefulness?
virtual cells · perturbation · single cell · OOD generalization · world modelsOpen frontier record →

Inputs and outputs

Inputs

Single-cell or perturbation data

Outputs

Cell statesPerturbation predictions

Scientific and technical profile

Scientific principles

Single-cell language modelingGenerative pretrainingTransfer learning

Technology

Gene/expression tokenizationMasked generative transformerFine-tunable heads
Ideas before algorithms

Scientific lineage

Explore all foundations

These are transparent concept matches—not claims that one scientist alone caused this model. Each connection is based on the model’s recorded domain, scientific principles, technical terms or an explicit lineage link.

Genomics & cell systems

Gene regulation and the operon model

François Jacob & Jacques Monod

Target biology, perturbation models, transcriptomic response prediction and virtual cells all require an explicit model of regulated gene programs.

Matched concepts: expression, perturbation, cell state
Computational intelligence

Transformer self-attention

Ashish Vaswani and colleagues

Protein, genome, molecule and single-cell foundation models use attention to learn dependencies across biological sequences and multimodal inputs.

Matched concepts: transformer, language model, foundation model
Computational intelligence

Information, entropy and communication

Claude E. Shannon

Sequence modelling, cross-entropy training, language models, mutual information and representation learning all use Shannon’s framework.

Matched concepts: language model, token
Genomics & cell systems

The epigenetic landscape and cell-fate trajectories

Conrad H. Waddington

Single-cell embeddings, trajectory inference, reprogramming and virtual-cell models often represent cell identity as movement through a learned state landscape.

Matched concepts: cell state, single-cell
Molecular recognition

Concerted allostery

Jacques Monod, Jeffries Wyman & Jean-Pierre Changeux

Allosteric drug design exploits remote pockets to modulate function, selectivity and resistance without competing at the active site.

Matched concepts: state
Biologics & genome engineering

Programmable CRISPR–Cas genome editing

Jennifer A. Doudna & Emmanuelle Charpentier

CRISPR enables target validation, disease models, perturbation atlases, functional genomics and gene-editing therapeutics.

Matched concepts: perturbation

Evaluation evidence

Dataset or evaluationSingle-cell downstream tasks
Task or metricCell representation
Evidence statusPeer-reviewed
Open source ↗

Task-specific evidence only; not comparable as a universal leaderboard score.

Single-cell representation learning

Single-cell downstream tasks

Version history not yet curated · Single-cell downstream-task benchmarks
peer-reviewed

Peer-reviewed evaluation across representation and downstream single-cell tasks.

Claim caveats
  • Performance varies by preprocessing, batch correction and downstream task.
  • Representation quality is not equivalent to causal perturbation prediction.

Known limitations

  • Performance depends on the evaluation dataset and operating conditions.
  • Task-specific benchmark results should not be compared across unlike domains.
  • Outputs require task-specific scientific and experimental validation.

Milestones

Not normalized

Pretrained on 33M+ human cells.

Evidence

One of the most-used open single-cell foundation models.