Model type
Study-specific predictive method; this record is the paper-specific evaluated configuration.
Configuration
C2S (GPT-2 Large)
Cell2Sentence adapts GPT-2 to single-cell transcriptomics by writing cells as expression-ranked gene-name sequences.
Sources
Cell2Sentence: Teaching Large Language Models the Language of Biology · Methods/Data transformation (paragraph 8); Introduction (paragraph 2)How it worksEvaluated procedure (conceptual)
Conceptual input–method–output guide. Check the procedure text and linked evaluation for fitted components, additional inputs and exact settings.
Sources
Cell2Sentence: Teaching Large Language Models the Language of Biology · Methods (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 3)At a glance
Biological inputs
Single-cell gene-expression profiles converted to ranked gene names
Sources
Cell2Sentence: Teaching Large Language Models the Language of Biology · Methods/Data transformation (paragraph 8); Methods/Data transformation (paragraph 3)Outputs
Cell-type annotations or generated ranked gene sequences
Sources
Cell2Sentence: Teaching Large Language Models the Language of Biology · Methods/Data transformation (paragraph 7); Inference Details (paragraph 2)limited source coverage · Automated source review, 2026-09-16. All specifications and missing details
Evaluations and results
Release 2026-09-17-d277315f7d76 · 1 evaluation · 1 metric row. Different protocols are not a single leaderboard.
| Metric and finding | Coverage and uncertainty | Evidence |
|---|---|---|
| C2S (GPT-2 Large): Combinatorial cell-label classification Partial-credit labels including cell type, perturbation, and dose. Author-reported evaluation · Evaluation metadata: needs review | ||
| 0.631 Partial-label accuracy Unit: unitless · Direction: unknown | Uncertainty: ± 0.0031 Scored: Not reported · Eligible: Not reported | source checkedCell2Sentence: Teaching Large Language Models the Language of Biology · Table 3, Partial label / C2S (GPT-2 Large) row, L1000 Acc column Source checking is not independent reproduction. |
How it works
How the evaluated method works
Genes are ordered by decreasing transcript abundance to create cell sentences. GPT-2 Large is fine-tuned on this representation for cell-type-conditioned generation and cell-type annotation.
Sources
Cell2Sentence: Teaching Large Language Models the Language of Biology · Methods (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 3)What was evaluated
The linked evaluation record identifies C2S (GPT-2 Large): Combinatorial cell-label classification. Its dataset, split, adaptation and evidence origin remain attached to the reported results.
Sources
Cell2Sentence: Teaching Large Language Models the Language of Biology · The named evaluation’s methods and comparison table; exact preserved evaluation IDs: evaluation-lit-027Strengths and limitations
Strengths and considerations
- Reuses text-language-model training tools for a biological representation that supports annotation and generation.
Sources
Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiments/Experiment 3: abstract summary generation/Objective: (paragraph 1); Abstract (paragraph 1)
Limitations and conditions
- Rank ordering does not directly retain measured expression magnitudes; reconstruction and generated-cell plausibility require the paper’s separate validation.
Sources
Cell2Sentence: Teaching Large Language Models the Language of Biology · Methods/Data transformation (paragraph 7); Inference Details (paragraph 2)
Profile review details
Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.
Stable record: reported-model-ab02228f50a37cSpecifications
Inputs, training, access and other details
Explanatory profile: limited source coverage · Automated source review, 2026-09-16. Review applies to the cited claims; unresolved fields are listed below. Numerical results retain their own review status.
| Property | Description and evidence |
|---|---|
| Model type | Study-specific predictive method; this record is the paper-specific evaluated configuration.SourcesCell2Sentence: Teaching Large Language Models the Language of Biology · Methods (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 3) |
| Architecture / procedure | Genes are ordered by decreasing transcript abundance to create cell sentences. GPT-2 Large is fine-tuned on this representation for cell-type-conditioned generation and cell-type annotation.SourcesCell2Sentence: Teaching Large Language Models the Language of Biology · Methods (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 3) |
| Biological inputs | Single-cell gene-expression profiles converted to ranked gene namesSourcesCell2Sentence: Teaching Large Language Models the Language of Biology · Methods/Data transformation (paragraph 8); Methods/Data transformation (paragraph 3) |
| Outputs | Cell-type annotations or generated ranked gene sequencesSourcesCell2Sentence: Teaching Large Language Models the Language of Biology · Methods/Data transformation (paragraph 7); Inference Details (paragraph 2) |
| Parameters | 774,030,080 parameters for the model checkpoint used in the Cell2Sentence L1000 experiment; this is not a claim about every checkpoint in the family.SourcesCell2Sentence: Teaching Large Language Models the Language of Biology · Table 5, Comparison of required compute on the L1000 dataset; # Parameters column and caption |
| Known versions / configuration | GPT-2 LargeSourcesCell2Sentence: Teaching Large Language Models the Language of Biology · Experiments/Experiment 3: abstract summary generation/Objective: (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 2) |
| Training data / fitting | Task-specific single-cell expression datasets described in the preprint; gene ranking replaces direct numeric expression input.SourcesCell2Sentence: Teaching Large Language Models the Language of Biology · Methods/Data transformation (paragraph 8); Experimental Details/Evaluation Datasets (paragraph 1) |
| Context limits | GPT-2 configurations use 1,024 tokens; the separately evaluated Pythia-160m configuration uses 9,200 tokens.SourcesCell2Sentence: Teaching Large Language Models the Language of Biology · Training Details (paragraph 1); Experiments (paragraph 1) |
| Access | Official study implementation and usage documentation: https://github.com/vandijklab/cell2sentence/blob/a6efaf079f98491d4723ced44b929936b94368aa/README.md. This pinned documentation revision is not automatically the evaluated weight revision.Sourcesvandijklab/cell2sentence README.md · README.md; installation, model download and usage instructions |
| Code licence | Apache 2.0 (study repository code at the cited revision; this does not establish every dependency or historical checkpoint licence).Sourcesvandijklab/cell2sentence LICENSE · LICENSE; complete licence text |
| Weights licence | The inspected model-access documentation does not explicitly identify terms for this exact evaluated checkpoint or fitted head; repository code terms are shown separately. · Not reported in inspected sourcesSourcesvandijklab/cell2sentence README.md · README.md; checkpoint/access documentation and licence scope |
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.
19 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|---|---|
| Diagram caption Conceptual input–method–output guide. Check the procedure text and linked evaluation for fitted components, additional inputs and exact settings. Individual claims | Cell2Sentence: Teaching Large Language Models the Language of Biology Methods (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 3) Version: preprint archived 2024-10-29 | source checked automated source review · 2026-09-16 Audit detailsPrimary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Diagram steps ["Single-cell gene-expression profiles converted to ranked gene names","C2S (GPT-2 Large)","Cell-type annotations or generated ranked gene sequences"] Individual claims | Cell2Sentence: Teaching Large Language Models the Language of Biology Methods (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 3) Version: preprint archived 2024-10-29 | source checked automated source review · 2026-09-16 Audit detailsPrimary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Diagram title Evaluated procedure (conceptual) Individual claims | Cell2Sentence: Teaching Large Language Models the Language of Biology Methods (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 3) Version: preprint archived 2024-10-29 | source checked automated source review · 2026-09-16 Audit detailsPrimary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Model type Study-specific predictive method; this record is the paper-specific evaluated configuration. Individual claims | Cell2Sentence: Teaching Large Language Models the Language of Biology Methods (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 3) Version: preprint archived 2024-10-29 | source checked automated source review · 2026-09-16 Audit detailsPrimary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Architecture / procedure Genes are ordered by decreasing transcript abundance to create cell sentences. GPT-2 Large is fine-tuned on this representation for cell-type-conditioned generation and cell-type annotation. Individual claims | Cell2Sentence: Teaching Large Language Models the Language of Biology Methods (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 3) Version: preprint archived 2024-10-29 | source checked automated source review · 2026-09-16 Audit detailsPrimary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Weights licence The inspected model-access documentation does not explicitly identify terms for this exact evaluated checkpoint or fitted head; repository code terms are shown separately. Individual claims | vandijklab/cell2sentence README.md README.md; checkpoint/access documentation and licence scope Version: a6efaf079f98491d4723ced44b929936b94368aa | unreported automated source review · 2026-09-16 Audit detailsPrimary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Biological inputs Single-cell gene-expression profiles converted to ranked gene names Individual claims | Cell2Sentence: Teaching Large Language Models the Language of Biology Methods/Data transformation (paragraph 8); Methods/Data transformation (paragraph 3) Version: preprint archived 2024-10-29 | source checked automated source review · 2026-09-16 Audit detailsPrimary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Outputs Cell-type annotations or generated ranked gene sequences Individual claims | Cell2Sentence: Teaching Large Language Models the Language of Biology Methods/Data transformation (paragraph 7); Inference Details (paragraph 2) Version: preprint archived 2024-10-29 | source checked automated source review · 2026-09-16 Audit detailsPrimary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Parameters 774,030,080 parameters for the model checkpoint used in the Cell2Sentence L1000 experiment; this is not a claim about every checkpoint in the family. Individual claims | Cell2Sentence: Teaching Large Language Models the Language of Biology Table 5, Comparison of required compute on the L1000 dataset; # Parameters column and caption Version: preprint archived 2024-10-29 | source checked automated source review · 2026-09-16 Audit detailsPrimary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
| Known versions / configuration GPT-2 Large Individual claims | Cell2Sentence: Teaching Large Language Models the Language of Biology Experiments/Experiment 3: abstract summary generation/Objective: (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 2) Version: preprint archived 2024-10-29 | source checked automated source review · 2026-09-16 Audit detailsPrimary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction. Field: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record |
Sources and history
Release 2026-09-17-d277315f7d76 · Record review: needs review
3 source records and release history
- Cell2Sentence: Teaching Large Language Models the Language of Biology · Original source · preprint archived 2024-10-29
- vandijklab/cell2sentence README.md · Original source · a6efaf079f98491d4723ced44b929936b94368aa
- vandijklab/cell2sentence LICENSE · Original source · a6efaf079f98491d4723ced44b929936b94368aa
Technical metadata and extraction receipts
Stable ID: reported-model-ab02228f50a37c
- areas
- cells-tissues
- entity level
- method
- version
- GPT-2 Large
- reported name
- C2S (GPT-2 Large)
- historical missing metadata
- checkpoint revision: not_reported_in_legacy_extract; training data: not_reported_in_legacy_extract; licence: not_reported_in_legacy_extract
- 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.
- legacy kinds
- model
- entity classification
- review date: 2026-09-17; rationale: This source-scoped entry preserves the method/configuration actually named in an evaluation. It is neither a global family identity nor proof of an immutable checkpoint; the linked evaluation retains adaptation, fitting and scoring details.; source ids: cell2sentence-2024; source locator: Methods (paragraph 1); Experiments/Fine-Tuning Datasets (paragraph 3) | Methods/Data transformation (paragraph 8); Introduction (paragraph 2); ambiguities: Configuration means the source-labelled evaluated identity. It does not establish missing checkpoint hashes, default settings or equivalence to same-named records in other papers.