Model

Geneformer

Geneformer represents single-cell transcriptomes as ranked genes and learns contextual gene and cell representations.

Sources (2)

ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter

5 evaluations · 5 metric rows · 1 evaluated configuration using this model

How it worksGeneformer workflow

Conceptual summary of the documented data flow; optional inputs and configured downstream stages must be reported for a reproducible evaluation.

Sources (2)

ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter

Overview

Model type

Masked-gene transcriptomic transformer encoder

Sources (2)

ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter

Inputs

Single-cell gene expression converted to corpus-normalized gene ranks.

Sources (2)

ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter

Outputs

Contextual gene and cell representations; task-specific outputs after the documented fine-tuning or perturbation workflow.

Sources (2)

ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter

Access

Official downloadable model/card and usage examples: https://huggingface.co/ctheodoris/Geneformer

Sources (2)

ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter

Source reviewed · Automated source review, 2026-09-16. All specifications and missing details

Evaluations and results

5 evaluations · 5 metric rows. Different protocols are not a single leaderboard.

Filter evaluations

Tested model or methodProtocolDatasetEvaluation setupMetricEvidence origin

Applied filters: All linked evaluations

Exact evaluated configurations and original reported results
Tested configuration	Protocol and dataset	Finding	Evidence and details
Configuration: Geneformer	Task: Cell-type identification Dataset: M.S. single-cell dataset	0.388 F1-Score unitless · unknown Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible	Independent external evaluation · source checked Methods, coverage and source Geneformer: Cell-type identification Native scLLM cell-type identification as reported in Table 2. Aggregation: Not reported Parameter-Efficient Fine-Tuning Enhances Adaptation of Single Cell Large Language Model for Cell Type Identification · Table 2, M.S. / Geneformer row, F1-Score column
Configuration: Geneformer	Task: Combinatorial cell-label classification Dataset: L1000	0.419 Partial-label accuracy unitless · unknown Uncertainty: ± 0.0153 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · source checked Methods, coverage and source Geneformer: Combinatorial cell-label classification Partial-credit labels including cell type, perturbation, and dose. Aggregation: Not reported Cell2Sentence: Teaching Large Language Models the Language of Biology · Table 3, Partial label / Geneformer row, L1000 Acc column
Configuration: Geneformer	Task: Cell-type annotation Dataset: hPancreas	0.270 F1 unitless · unknown Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible	Result quoted from another source · source checked Methods, coverage and source Geneformer: Cell-type annotation Zero-shot setting; source caption says some comparator rows come from GenePT. Aggregation: Not reported scELMo: Embeddings from Language Models are Good Learners for Single-cell Data Analysis · Table 1, hPancreas zero-shot / Geneformer (z) row, F1 column
Configuration: Mouse-Geneformer	Task: Human thymus cell-type classification Dataset: Human thymus scRNA-seq	48.57 F1 % · unknown Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · source checked Methods, coverage and source Mouse-Geneformer: Human thymus cell-type classification Ortholog-based gene conversion; zero-shot mouse model on human cells. Aggregation: Not reported Mouse-Geneformer: A deep learning model for mouse single-cell transcriptome and its cross-species utility · Table 4, h/ Thymus row, Mouse-Geneformer Zero-shot F1 column
Configuration: Human-Geneformer	Task: Human thymus cell-type classification Dataset: Human thymus scRNA-seq	74.48 F1 % · unknown Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible	Independent external evaluation · source checked Methods, coverage and source Human-Geneformer: Human thymus cell-type classification Native human model; zero-shot setting. Aggregation: Not reported Mouse-Geneformer: A deep learning model for mouse single-cell transcriptome and its cross-species utility · Table 4, h/ Thymus row, Human-Geneformer Zero-shot F1 column

Source checking is not independent reproduction. Release 2026-09-23-2b89723c6dd9.

Related configurations, pipelines and services

These configurations, services and pipelines use this model within their own configurations. Their results, where available, are not assigned to the underlying model.

scRegNet (Geneformer backbone) Pipeline · 1 metric rows

Use this model

How it works, versions and access

Versions and evaluated configurations

Geneformer · family
Human-Geneformer · variant of
Geneformer · family
Geneformer · family
Mouse-Geneformer · variant of

How it works

Geneformer represents single-cell transcriptomes as ranked genes and learns contextual gene and cell representations. Transformer encoder trained to recover masked genes from rank-value-encoded expression profiles. The documented inputs are single-cell gene expression converted to corpus-normalized gene ranks. The output consists of contextual gene and cell representations; task-specific outputs after the documented fine-tuning or perturbation workflow.

Sources (2)

ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter

Versions and reproducibility

Geneformer-V1-10M, V2-104M, V2-316M and V2-104M_CLcancer; the card states V2-316M is the repository default. V1: 2,048 gene tokens; V2: 4,096. Vocabularies also differ.

Sources (2)

ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter

Strengths, limitations and unresolved questions

Strengths and limitations

Strengths supported by sources

Self-supervised pretraining uses unlabeled cells and supports downstream gene/cell classification and in-silico perturbation workflows.
Sources (2)
ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter

Limitations and conditions

V1, V2 and cancer-tuned V2 use different corpora and vocabularies. The authors recommend task-specific hyperparameter tuning; there is no universally suitable fine-tuning configuration.
Sources (2)
ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter

Profile review details

Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied. Follow-up retrieved the complete original RNA-FM PDF and inspected the full pinned Geneformer repository inventory; unavailable labels were updated only where new evidence resolved the earlier retrieval gap.

Stable record: catalog-model-geneformer

Specifications

Inputs, training, access and other details

Explanatory profile: source reviewed · Automated source review, 2026-09-16. Review applies to the cited claims; unresolved fields are listed below. Numerical results retain their own review status.

Inputs, outputs and configuration
Property	Description and evidence
Model type	Masked-gene transcriptomic transformer encoder Sources (2) ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter
Architecture	Transformer encoder trained to recover masked genes from rank-value-encoded expression profiles. Sources (2) ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter
Inputs	Single-cell gene expression converted to corpus-normalized gene ranks. Sources (2) ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter
Outputs	Contextual gene and cell representations; task-specific outputs after the documented fine-tuning or perturbation workflow. Sources (2) ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter
Parameters	V1: 10M; V2: 104M or 316M. Sources (2) ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter
Known versions	Geneformer-V1-10M, V2-104M, V2-316M and V2-104M_CLcancer; the card states V2-316M is the repository default. Sources (2) ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter
Training data	V1: approximately 30M human single-cell transcriptomes. V2: approximately 104M non-cancer human transcriptomes; the cancer continual-learning variant adds approximately 14M cancer cells. Sources (2) ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter
Training cutoff	Training dates reported: June 2021 for V1 and December 2024 for V2. These are training dates, not independently verified data-collection cutoffs. Sources (2) ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter
Context limits	V1: 2,048 gene tokens; V2: 4,096. Vocabularies also differ. Sources (2) ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter
Weights licence	Apache-2.0 is declared in the official model-card metadata. Sources (2) ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter
Access	Official downloadable model/card and usage examples: https://huggingface.co/ctheodoris/Geneformer Sources (2) ctheodoris/Geneformer: README.md; ctheodoris/Geneformer: config.json · README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter
Code licence	The official repository declares Apache-2.0 in its model card. No separate code-licence file appears in the complete inspected revision; this records the repository declaration rather than an independently reviewed licence grant for every bundled dependency. Sources (2) ctheodoris/Geneformer: README.md; Geneformer official repository file inventory · Pinned README license metadata; complete two-page recursive repository inventory at revision 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5

Evidence

Source checking verifies the cited claim or transcription. It does not establish independent reproduction.

Evidence table

Inspect claims, sources and review details

Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.

One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.

40 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-09-23-2b89723c6dd9
Property and statement	Original source and location	Review and provenance
Diagram caption Conceptual summary of the documented data flow; optional inputs and configured downstream stages must be reported for a reproducible evaluation. Individual claims	ctheodoris/Geneformer: README.md Original source ↗ README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5 Retrieved: 2026-09-16T19:46:20.640731+00:00	source checked automated source review · 2026-09-16 Audit details Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied. Follow-up retrieved the complete original RNA-FM PDF and inspected the full pinned Geneformer repository inventory; unavailable labels were updated only where new evidence resolved the earlier retrieval gap. Field: `attributes.profile.diagram.caption` Source artifact SHA-256: `56d6e570b349cbedae9a54634421c94e7af8ea467ce0fdd79193372ae3cbdbd8` Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact Inspected artifact
Diagram caption Conceptual summary of the documented data flow; optional inputs and configured downstream stages must be reported for a reproducible evaluation. Individual claims	ctheodoris/Geneformer: config.json Original source ↗ README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5 Retrieved: 2026-09-16T19:46:20.640731+00:00	source checked automated source review · 2026-09-16 Audit details Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied. Follow-up retrieved the complete original RNA-FM PDF and inspected the full pinned Geneformer repository inventory; unavailable labels were updated only where new evidence resolved the earlier retrieval gap. Field: `attributes.profile.diagram.caption` Source artifact SHA-256: `2cc4af3442644e84af71814a607b61c958d7b4e5ebde6791ab77b5f534ac6f6e` Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact Inspected artifact
Diagram steps Expression profile Corpus-normalized gene ranks Masked-gene transformer Cell and gene representations Individual claims	ctheodoris/Geneformer: README.md Original source ↗ README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5 Retrieved: 2026-09-16T19:46:20.640731+00:00	source checked automated source review · 2026-09-16 Audit details Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied. Follow-up retrieved the complete original RNA-FM PDF and inspected the full pinned Geneformer repository inventory; unavailable labels were updated only where new evidence resolved the earlier retrieval gap. Field: `attributes.profile.diagram.steps` Source artifact SHA-256: `56d6e570b349cbedae9a54634421c94e7af8ea467ce0fdd79193372ae3cbdbd8` Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact Inspected artifact
Diagram steps Expression profile Corpus-normalized gene ranks Masked-gene transformer Cell and gene representations Individual claims	ctheodoris/Geneformer: config.json Original source ↗ README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5 Retrieved: 2026-09-16T19:46:20.640731+00:00	source checked automated source review · 2026-09-16 Audit details Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied. Follow-up retrieved the complete original RNA-FM PDF and inspected the full pinned Geneformer repository inventory; unavailable labels were updated only where new evidence resolved the earlier retrieval gap. Field: `attributes.profile.diagram.steps` Source artifact SHA-256: `2cc4af3442644e84af71814a607b61c958d7b4e5ebde6791ab77b5f534ac6f6e` Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact Inspected artifact
Diagram title Geneformer workflow Individual claims	ctheodoris/Geneformer: README.md Original source ↗ README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5 Retrieved: 2026-09-16T19:46:20.640731+00:00	source checked automated source review · 2026-09-16 Audit details Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied. Follow-up retrieved the complete original RNA-FM PDF and inspected the full pinned Geneformer repository inventory; unavailable labels were updated only where new evidence resolved the earlier retrieval gap. Field: `attributes.profile.diagram.title` Source artifact SHA-256: `56d6e570b349cbedae9a54634421c94e7af8ea467ce0fdd79193372ae3cbdbd8` Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact Inspected artifact
Diagram title Geneformer workflow Individual claims	ctheodoris/Geneformer: config.json Original source ↗ README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5 Retrieved: 2026-09-16T19:46:20.640731+00:00	source checked automated source review · 2026-09-16 Audit details Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied. Follow-up retrieved the complete original RNA-FM PDF and inspected the full pinned Geneformer repository inventory; unavailable labels were updated only where new evidence resolved the earlier retrieval gap. Field: `attributes.profile.diagram.title` Source artifact SHA-256: `2cc4af3442644e84af71814a607b61c958d7b4e5ebde6791ab77b5f534ac6f6e` Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact Inspected artifact
Model type Masked-gene transcriptomic transformer encoder Individual claims	ctheodoris/Geneformer: README.md Original source ↗ README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5 Retrieved: 2026-09-16T19:46:20.640731+00:00	source checked automated source review · 2026-09-16 Audit details Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied. Follow-up retrieved the complete original RNA-FM PDF and inspected the full pinned Geneformer repository inventory; unavailable labels were updated only where new evidence resolved the earlier retrieval gap. Field: `attributes.profile.facts.0.value` Source artifact SHA-256: `56d6e570b349cbedae9a54634421c94e7af8ea467ce0fdd79193372ae3cbdbd8` Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact Inspected artifact
Model type Masked-gene transcriptomic transformer encoder Individual claims	ctheodoris/Geneformer: config.json Original source ↗ README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5 Retrieved: 2026-09-16T19:46:20.640731+00:00	source checked automated source review · 2026-09-16 Audit details Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied. Follow-up retrieved the complete original RNA-FM PDF and inspected the full pinned Geneformer repository inventory; unavailable labels were updated only where new evidence resolved the earlier retrieval gap. Field: `attributes.profile.facts.0.value` Source artifact SHA-256: `2cc4af3442644e84af71814a607b61c958d7b4e5ebde6791ab77b5f534ac6f6e` Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact Inspected artifact
Architecture Transformer encoder trained to recover masked genes from rank-value-encoded expression profiles. Individual claims	ctheodoris/Geneformer: README.md Original source ↗ README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5 Retrieved: 2026-09-16T19:46:20.640731+00:00	source checked automated source review · 2026-09-16 Audit details Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied. Follow-up retrieved the complete original RNA-FM PDF and inspected the full pinned Geneformer repository inventory; unavailable labels were updated only where new evidence resolved the earlier retrieval gap. Field: `attributes.profile.facts.1.value` Source artifact SHA-256: `56d6e570b349cbedae9a54634421c94e7af8ea467ce0fdd79193372ae3cbdbd8` Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact Inspected artifact
Architecture Transformer encoder trained to recover masked genes from rank-value-encoded expression profiles. Individual claims	ctheodoris/Geneformer: config.json Original source ↗ README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5 Retrieved: 2026-09-16T19:46:20.640731+00:00	source checked automated source review · 2026-09-16 Audit details Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied. Follow-up retrieved the complete original RNA-FM PDF and inspected the full pinned Geneformer repository inventory; unavailable labels were updated only where new evidence resolved the earlier retrieval gap. Field: `attributes.profile.facts.1.value` Source artifact SHA-256: `2cc4af3442644e84af71814a607b61c958d7b4e5ebde6791ab77b5f534ac6f6e` Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact Inspected artifact

All evidence (CSV)All evidence (JSONL)How to read the evidence

Sources and history

View linked audit checks and correction history

Release 2026-09-23-2b89723c6dd9 · Record review: discovered

4 source records and release history

Geneformer official resource · Original source · Not reported
ctheodoris/Geneformer: README.md · Original source · 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5
ctheodoris/Geneformer: config.json · Original source · 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5
Geneformer official repository file inventory · Original source · 1f7fbae4e469a5f4f1af8c111a529cfe1b3829f5

Download this release

Technical metadata and extraction receipts

Stable ID: catalog-model-geneformer

areas: cells-tissues
method types: foundation model
entity level: family
version: published checkpoints
reported name: Geneformer
access: Public checkpoints; specify exact version before evaluation.
method type: foundation model
historical missing metadata: checkpoint revision: not_yet_extracted; training data: not_yet_extracted; licence: not_yet_extracted
metadata review scope: historical_missing_metadata preserves the original discovery state. Current descriptive evidence and missingness are recorded in profile.facts; numerical-result review is separate.
entity classification: review date: 2026-09-17; rationale: The cited profile describes a named learned biological predictor or representation model/family. Preserve this identity separately from task-specific fitting, individual checkpoints, pipelines and hosted access.; source ids: evidence-official-d3d9afa5c6172982ed01; evidence-official-f60303c5eaa60ed9c77a; source locator: README.md: Model Description, pretrained model list, Application, Installation and licence frontmatter; ambiguities: None recorded

Related records

family: Geneformer
variant of: Human-Geneformer
uses model: scRegNet (Geneformer backbone)
family: Geneformer
family: Geneformer
variant of: Mouse-Geneformer

Suggest a correction