Task

Combinatorial cell-label classification

Combinatorial cell-label classification tests exact labels and partial correctness across single-cell and bulk-expression datasets.

Sources

Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113

2 evaluations · 2 results

Overview

Datasets

Cytokine stimulation, L1000 and GTEx datasets with multipart metadata labels.

Metrics

Accuracy and AUROC are reported separately for exact full-label matching and partial-label credit.

Allowed inputs

Gene-expression-derived cell representations.

Sources

Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113

Evaluation procedure diagram

How it worksComputational evaluation flow

Conceptual summary of the cited evaluation; exact task configuration and source version remain part of the protocol.

Sources

Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113

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

Results

Results are available, but no reviewed comparison panel is linked in this release.

All evaluations

2 evaluations · 2 results. 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: C2S (GPT-2 Large)	Task: Combinatorial cell-label classification Dataset: L1000	0.631 Partial-label accuracy unitless · unknown Uncertainty: ± 0.0031 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Methods, coverage and source C2S (GPT-2 Large): 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 / C2S (GPT-2 Large) row, L1000 Acc 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

Source checking is not independent reproduction. Release 2026-09-29-06401fd5b220.

Methods and evaluation design

Procedure, tasks and evaluated configurations

How it works

Evaluation methodology

Cytokine stimulation, L1000 and GTEx datasets with multipart metadata labels. The cytokine label-classification setting includes all label combinations during training; held-out combinations belong to a separate generation task. Accuracy and AUROC are reported separately for exact full-label matching and partial-label credit. k-nearest-neighbour, XGBoost, Geneformer and scGPT comparators. For combinatorial label classification, all combinations of cytokine perturbations are included in training. Holding out 10 of 140 cytokine combinations belongs to the separate perturbed-cell-generation experiment, not this classifier. L1000 and GTEx provide bulk-expression evaluation outside the single-cell fine-tuning distribution.

Sources

Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113; Fine-Tuning Datasets; Experiment 2: combinatorial label classification

Recorded evaluations

Each evaluation records what was tested and under which conditions.

Run instructions

No runnable recipe has been reviewed for this task. Dataset access, model requirements, licences and compute requirements must be checked against its sources before execution.

A task describes a biological question. Choose a linked protocol to obtain concrete split and scoring instructions.

Strengths, limitations and unresolved questions

Strengths and limitations

Strengths supported by sources

No source-reviewed explanatory claims are recorded here yet.

Limitations and conditions

The combinatorial classifier includes every cytokine-label combination during training. A held-out-combination claim belongs to the separate generation task; exact and partial label credit are also different metrics.
Sources
Cell2Sentence: Teaching Large Language Models the Language of Biology · Fine-Tuning Datasets; Experiment 2: combinatorial label classification; Experiments: Experiment 2/cell label prediction/Methodology; Fine-Tuning Datasets

Profile review details

Targeted full-paper and supplement review of the outstanding task fields, with original dataset metadata checked where accessible. Source-scoped omissions are explicit; no independent benchmark reproduction or numerical-result change.

Stable record: reported-task-7efe245cc94ee5

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.

Data, procedure and scoring
Property	Description and evidence
Datasets	Cytokine stimulation, L1000 and GTEx datasets with multipart metadata labels. Sources Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113
Splits	The cytokine label-classification setting includes all label combinations during training; held-out combinations belong to a separate generation task. Sources Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113
Metrics	Accuracy and AUROC are reported separately for exact full-label matching and partial-label credit. Sources Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113
Baselines	k-nearest-neighbour, XGBoost, Geneformer and scGPT comparators. Sources Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113
Leakage controls	For combinatorial label classification, all combinations of cytokine perturbations are included in training. Holding out 10 of 140 cytokine combinations belongs to the separate perturbed-cell-generation experiment, not this classifier. L1000 and GTEx provide bulk-expression evaluation outside the single-cell fine-tuning distribution. Sources Cell2Sentence: Teaching Large Language Models the Language of Biology · Fine-Tuning Datasets; Experiment 2: combinatorial label classification
Uncertainty	Three experimental repeats are reported for each dataset. Sources Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113
Entity type	Paper-specific computational evaluation protocol. Sources Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113
Organisms	The cytokine-stimulation arm explicitly uses human peripheral blood mononuclear cells. The two additional arms use L1000 and GTEx bulk-expression collections, with their cell-line and tissue identities retained in combinatorial labels rather than treated as the same PBMC population. Sources Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiments: Experiment 2/cell label prediction/Methodology; Fine-Tuning Datasets
Assays	Expression measurements with multipart cell/tissue/stimulation labels. Sources Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113
Allowed inputs	Gene-expression-derived cell representations. Sources Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113
Adaptation	Supervised label classification includes all label combinations in training; compositional generation is a different experiment. Sources Cell2Sentence: Teaching Large Language Models the Language of Biology · Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113

Evidence

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

Papers and result coverage

Last literature check: 2026-09-17. Primary-paper discovery and source inspection. Source-checked results are not independently reproduced experiments.

Paper or primary resource	Version	Reference
Cell2Sentence: Teaching Large Language Models the Language of Biology	preprint archived 2024-10-29	Read source DOI: 10.1101/2023.09.11.557287

Historical gaps recorded on 2026-09-17

The catalogue now holds 2 result rows for this benchmark. A note below about pending extraction describes the state on 2026-09-17 and may since have been answered by a later batch. The result rows and their sources are the current record.

Partial-credit and full-label metrics cannot share a ranking; keep dataset and label regime explicit.
Version is the archived preprint, not a later model family release.

Search and extraction details

primary comparison table screened

Searches

"PMC11565894"

Evidence locations

Table 3
Fine-Tuning Datasets
Classification task methodology

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.

17 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-09-29-06401fd5b220
Property and statement	Original source and location	Review and provenance
Diagram caption Conceptual summary of the cited evaluation; exact task configuration and source version remain part of the protocol. Individual claims	Cell2Sentence: Teaching Large Language Models the Language of Biology Original source ↗ Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113 Version: preprint archived 2024-10-29 Retrieved: 2026-09-16T10:41:16.533640+00:00	source checked automated source review · 2026-09-16 Audit details Targeted full-paper and supplement review of the outstanding task fields, with original dataset metadata checked where accessible. Source-scoped omissions are explicit; no independent benchmark reproduction or numerical-result change. Field: `attributes.profile.diagram.caption` Source artifact SHA-256: `e088727d6e04857fccb7033a9b074e1850f775e86e7d2e99e603dde09558ab02` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Diagram steps Input: Gene-expression-derived cell representations. Evaluation: The cytokine label-classification setting includes all label combinations during training; held-out combinations belong to a separate generation task. Readout: Accuracy and AUROC are reported separately for exact full-label matching and partial-label credit. Individual claims	Cell2Sentence: Teaching Large Language Models the Language of Biology Original source ↗ Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113 Version: preprint archived 2024-10-29 Retrieved: 2026-09-16T10:41:16.533640+00:00	source checked automated source review · 2026-09-16 Audit details Targeted full-paper and supplement review of the outstanding task fields, with original dataset metadata checked where accessible. Source-scoped omissions are explicit; no independent benchmark reproduction or numerical-result change. Field: `attributes.profile.diagram.steps` Source artifact SHA-256: `e088727d6e04857fccb7033a9b074e1850f775e86e7d2e99e603dde09558ab02` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Diagram title Computational evaluation flow Individual claims	Cell2Sentence: Teaching Large Language Models the Language of Biology Original source ↗ Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113 Version: preprint archived 2024-10-29 Retrieved: 2026-09-16T10:41:16.533640+00:00	source checked automated source review · 2026-09-16 Audit details Targeted full-paper and supplement review of the outstanding task fields, with original dataset metadata checked where accessible. Source-scoped omissions are explicit; no independent benchmark reproduction or numerical-result change. Field: `attributes.profile.diagram.title` Source artifact SHA-256: `e088727d6e04857fccb7033a9b074e1850f775e86e7d2e99e603dde09558ab02` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Datasets Cytokine stimulation, L1000 and GTEx datasets with multipart metadata labels. Individual claims	Cell2Sentence: Teaching Large Language Models the Language of Biology Original source ↗ Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113 Version: preprint archived 2024-10-29 Retrieved: 2026-09-16T10:41:16.533640+00:00	source checked automated source review · 2026-09-16 Audit details Targeted full-paper and supplement review of the outstanding task fields, with original dataset metadata checked where accessible. Source-scoped omissions are explicit; no independent benchmark reproduction or numerical-result change. Field: `attributes.profile.facts.0.value` Source artifact SHA-256: `e088727d6e04857fccb7033a9b074e1850f775e86e7d2e99e603dde09558ab02` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Splits The cytokine label-classification setting includes all label combinations during training; held-out combinations belong to a separate generation task. Individual claims	Cell2Sentence: Teaching Large Language Models the Language of Biology Original source ↗ Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113 Version: preprint archived 2024-10-29 Retrieved: 2026-09-16T10:41:16.533640+00:00	source checked automated source review · 2026-09-16 Audit details Targeted full-paper and supplement review of the outstanding task fields, with original dataset metadata checked where accessible. Source-scoped omissions are explicit; no independent benchmark reproduction or numerical-result change. Field: `attributes.profile.facts.1.value` Source artifact SHA-256: `e088727d6e04857fccb7033a9b074e1850f775e86e7d2e99e603dde09558ab02` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Adaptation Supervised label classification includes all label combinations in training; compositional generation is a different experiment. Individual claims	Cell2Sentence: Teaching Large Language Models the Language of Biology Original source ↗ Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113 Version: preprint archived 2024-10-29 Retrieved: 2026-09-16T10:41:16.533640+00:00	source checked automated source review · 2026-09-16 Audit details Targeted full-paper and supplement review of the outstanding task fields, with original dataset metadata checked where accessible. Source-scoped omissions are explicit; no independent benchmark reproduction or numerical-result change. Field: `attributes.profile.facts.10.value` Source artifact SHA-256: `e088727d6e04857fccb7033a9b074e1850f775e86e7d2e99e603dde09558ab02` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Metrics Accuracy and AUROC are reported separately for exact full-label matching and partial-label credit. Individual claims	Cell2Sentence: Teaching Large Language Models the Language of Biology Original source ↗ Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113 Version: preprint archived 2024-10-29 Retrieved: 2026-09-16T10:41:16.533640+00:00	source checked automated source review · 2026-09-16 Audit details Targeted full-paper and supplement review of the outstanding task fields, with original dataset metadata checked where accessible. Source-scoped omissions are explicit; no independent benchmark reproduction or numerical-result change. Field: `attributes.profile.facts.2.value` Source artifact SHA-256: `e088727d6e04857fccb7033a9b074e1850f775e86e7d2e99e603dde09558ab02` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Baselines k-nearest-neighbour, XGBoost, Geneformer and scGPT comparators. Individual claims	Cell2Sentence: Teaching Large Language Models the Language of Biology Original source ↗ Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113 Version: preprint archived 2024-10-29 Retrieved: 2026-09-16T10:41:16.533640+00:00	source checked automated source review · 2026-09-16 Audit details Targeted full-paper and supplement review of the outstanding task fields, with original dataset metadata checked where accessible. Source-scoped omissions are explicit; no independent benchmark reproduction or numerical-result change. Field: `attributes.profile.facts.3.value` Source artifact SHA-256: `e088727d6e04857fccb7033a9b074e1850f775e86e7d2e99e603dde09558ab02` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Leakage controls For combinatorial label classification, all combinations of cytokine perturbations are included in training. Holding out 10 of 140 cytokine combinations belongs to the separate perturbed-cell-generation experiment, not this classifier. L1000 and GTEx provide bulk-expression evaluation outside the single-cell fine-tuning distribution. Individual claims	Cell2Sentence: Teaching Large Language Models the Language of Biology Original source ↗ Fine-Tuning Datasets; Experiment 2: combinatorial label classification Version: preprint archived 2024-10-29 Retrieved: 2026-09-16T10:41:16.533640+00:00	source checked automated source review · 2026-09-16 Audit details Targeted full-paper and supplement review of the outstanding task fields, with original dataset metadata checked where accessible. Source-scoped omissions are explicit; no independent benchmark reproduction or numerical-result change. Field: `attributes.profile.facts.4.value` Source artifact SHA-256: `e088727d6e04857fccb7033a9b074e1850f775e86e7d2e99e603dde09558ab02` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Uncertainty Three experimental repeats are reported for each dataset. Individual claims	Cell2Sentence: Teaching Large Language Models the Language of Biology Original source ↗ Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113 Version: preprint archived 2024-10-29 Retrieved: 2026-09-16T10:41:16.533640+00:00	source checked automated source review · 2026-09-16 Audit details Targeted full-paper and supplement review of the outstanding task fields, with original dataset metadata checked where accessible. Source-scoped omissions are explicit; no independent benchmark reproduction or numerical-result change. Field: `attributes.profile.facts.5.value` Source artifact SHA-256: `e088727d6e04857fccb7033a9b074e1850f775e86e7d2e99e603dde09558ab02` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact

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

Sources and history

View linked audit checks and correction history

Release 2026-09-29-06401fd5b220 · Record review: needs review

2 source records and release history

Cell2Sentence: Teaching Large Language Models the Language of Biology · Original source · preprint archived 2024-10-29
Cell2Sentence: Teaching Large Language Models the Language of Biology · Original source · preprint archived 2024-10-29

Download this release

Technical metadata and extraction receipts

Stable ID: reported-task-7efe245cc94ee5

areas: cells-tissues
tasks: Combinatorial cell-label classification
entity level: task
version: Not reported
task: Combinatorial cell-label classification
scope note: Paper-specific evaluation task; protocol completeness requires further extraction.
benchmark research: review date: 2026-09-17; status: primary_comparison_table_screened; primary sources: expansion-p3-cell2sentence-2024; inspected locators: Table 3; Fine-Tuning Datasets; Classification task methodology; searched queries: "PMC11565894"; gaps: Partial-credit and full-label metrics cannot share a ranking; keep dataset and label regime explicit.; Version is the archived preprint, not a later model family release.; claim scope: Primary-paper discovery and source inspection. Source-checked results are not independently reproduced experiments.
historical missing metadata: protocol version: not_reported_in_legacy_extract; split: 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: benchmark
entity classification: review date: 2026-09-17; rationale: This source-scoped record identifies the biological prediction task and holds its paper context. Preserve the existing task identity; exact split, model adaptation and scoring remain in linked evaluations or separate protocol records.; source ids: cell2sentence-2024; source locator: Experiment 2: cell label classification; Table 3; Fine-Tuning Datasets; cached text lines 47–53, 77–78, 113; ambiguities: A paper- or suite-specific task may constrain some inputs or metrics; that alone does not make it interchangeable with a complete versioned protocol. No protocol equivalence is inferred.; Some legacy profile Entity type facts use the generic phrase computational evaluation protocol. That boilerplate is not sufficient to establish a single fixed protocol identity or to merge this task with another protocol record.

Related records

dataset: L1000
benchmark: C2S (GPT-2 Large): Combinatorial cell-label classification
benchmark: Geneformer: Combinatorial cell-label classification

Suggest a correction