Protocol

Chen et al. 2026 supervised cell type annotation, ATAA

Supervised fine-tuning on reference labels with a stratified 8:2 train/test split; metrics on the shared held-out test set. Dataset ATAA.

5 evaluations · 30 results

Overview

An explanatory profile has not yet been reviewed for this record.

Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.

5 recorded evaluations, 30 metric rows. A comparison chart has not yet been validated for these results. The table retains the individual findings and their sources.

View coverage and remaining gaps across all benchmarks

Results

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

All evaluations

5 evaluations · 30 results. Different protocols are not a single leaderboard.

Protocol: Chen et al. 2026 supervised cell type annotation, ATAA · Dataset subset: ATAA (Chen et al. 2026)

Sorted by Accuracy (held-out test set) (higher is better). The best value in each column is highlighted. Decimals are rounded for display; each value links to the printed value and its source.

Tested configuration	Accuracy (held-out test set)	AUPRC (held-out test set; macro-averaged over classes)	AUROC (held-out test set; macro-averaged over classes)	Macro F1 (held-out test set)	Precision (held-out test set; described as global in the Methods, averaging not further defined)	Recall (held-out test set; described as global in the Methods, averaging not further defined)
CellPLM (fine-tuned, Chen et al. 2026)Evaluation	0.97 (best in column)	0.98 (best in column)	0.99 (best in column)	0.95 (best in column)	0.97 (best in column)	0.97 (best in column)
Nicheformer (fine-tuned, Chen et al. 2026)Evaluation	0.97 (best in column)	0.98 (best in column)	0.99 (best in column)	0.94	0.97 (best in column)	0.97 (best in column)
scGPT-spatial (fine-tuned, Chen et al. 2026)Evaluation	0.97 (best in column)	0.98 (best in column)	0.99 (best in column)	0.94	0.95	0.94
scELMo (fine-tuned, Chen et al. 2026)Evaluation	0.95	0.95	0.99 (best in column)	0.89	0.96	0.86
GenePT-w (fine-tuned, Chen et al. 2026)Evaluation	0.92	0.87	0.96	0.85	0.92	0.81

All 30 result rows with coverage, uncertainty and sources

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: CellPLM (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.97 Accuracy (held-out test set) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'CellPLM', dataset 'ATAA', column 'Accuracy'
Configuration: CellPLM (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.95 Macro F1 (held-out test set) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'CellPLM', dataset 'ATAA', column 'Macro-F1'
Configuration: CellPLM (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.98 AUPRC (held-out test set; macro-averaged over classes) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'CellPLM', dataset 'ATAA', column 'PR-AUC'
Configuration: CellPLM (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.97 Precision (held-out test set; described as global in the Methods, averaging not further defined) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'CellPLM', dataset 'ATAA', column 'Precision'
Configuration: CellPLM (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.97 Recall (held-out test set; described as global in the Methods, averaging not further defined) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'CellPLM', dataset 'ATAA', column 'Recall'
Configuration: CellPLM (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.99 AUROC (held-out test set; macro-averaged over classes) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'CellPLM', dataset 'ATAA', column 'ROC-AUC'
Configuration: GenePT-w (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.92 Accuracy (held-out test set) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source GenePT-w on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'GenePT-w', dataset 'ATAA', column 'Accuracy'
Configuration: GenePT-w (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.85 Macro F1 (held-out test set) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source GenePT-w on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'GenePT-w', dataset 'ATAA', column 'Macro-F1'
Configuration: GenePT-w (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.87 AUPRC (held-out test set; macro-averaged over classes) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source GenePT-w on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'GenePT-w', dataset 'ATAA', column 'PR-AUC'
Configuration: GenePT-w (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.92 Precision (held-out test set; described as global in the Methods, averaging not further defined) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source GenePT-w on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'GenePT-w', dataset 'ATAA', column 'Precision'
Configuration: GenePT-w (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.81 Recall (held-out test set; described as global in the Methods, averaging not further defined) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source GenePT-w on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'GenePT-w', dataset 'ATAA', column 'Recall'
Configuration: GenePT-w (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.96 AUROC (held-out test set; macro-averaged over classes) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source GenePT-w on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'GenePT-w', dataset 'ATAA', column 'ROC-AUC'
Configuration: Nicheformer (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.97 Accuracy (held-out test set) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source Nicheformer on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'Nicheformer', dataset 'ATAA', column 'Accuracy'
Configuration: Nicheformer (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.94 Macro F1 (held-out test set) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source Nicheformer on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'Nicheformer', dataset 'ATAA', column 'Macro-F1'
Configuration: Nicheformer (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.98 AUPRC (held-out test set; macro-averaged over classes) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source Nicheformer on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'Nicheformer', dataset 'ATAA', column 'PR-AUC'
Configuration: Nicheformer (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.97 Precision (held-out test set; described as global in the Methods, averaging not further defined) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source Nicheformer on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'Nicheformer', dataset 'ATAA', column 'Precision'
Configuration: Nicheformer (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.97 Recall (held-out test set; described as global in the Methods, averaging not further defined) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source Nicheformer on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'Nicheformer', dataset 'ATAA', column 'Recall'
Configuration: Nicheformer (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.99 AUROC (held-out test set; macro-averaged over classes) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source Nicheformer on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'Nicheformer', dataset 'ATAA', column 'ROC-AUC'
Configuration: scELMo (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.95 Accuracy (held-out test set) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source scELMo on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'scELMo', dataset 'ATAA', column 'Accuracy'
Configuration: scELMo (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.89 Macro F1 (held-out test set) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source scELMo on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'scELMo', dataset 'ATAA', column 'Macro-F1'
Configuration: scELMo (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.95 AUPRC (held-out test set; macro-averaged over classes) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source scELMo on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'scELMo', dataset 'ATAA', column 'PR-AUC'
Configuration: scELMo (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.96 Precision (held-out test set; described as global in the Methods, averaging not further defined) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source scELMo on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'scELMo', dataset 'ATAA', column 'Precision'
Configuration: scELMo (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.86 Recall (held-out test set; described as global in the Methods, averaging not further defined) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source scELMo on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'scELMo', dataset 'ATAA', column 'Recall'
Configuration: scELMo (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.99 AUROC (held-out test set; macro-averaged over classes) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source scELMo on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'scELMo', dataset 'ATAA', column 'ROC-AUC'
Configuration: scGPT-spatial (fine-tuned, Chen et al. 2026)	Protocol: Chen et al. 2026 supervised cell type annotation, ATAA Dataset subset: ATAA (Chen et al. 2026)	0.97 Accuracy (held-out test set) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 4 precision and recall equal accuracy in all eight Nicheformer and CellPLM rows, as micro (global) averaging would give, but recall differs from accuracy in every scGPT-spatial, GenePT-w, scELMo and Novae row. The Methods call precision and recall 'global' without defining the averaging, so they may be averaged differently between models and should not be compared across models.Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) Methods, coverage and source scGPT-spatial on Chen et al. 2026 supervised cell type annotation, ATAA lit-cells-networks-20261011-chen2026-protocol-annotation-ataa Aggregation: Not reported Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Table 4, model 'scGPT-spatial', dataset 'ATAA', column 'Accuracy'

Source checking is not independent reproduction. Release 2026-10-23-5347695500ed.

Methods and evaluation design

Procedure, tasks and evaluated configurations

Recorded evaluations

Each evaluation records what was tested and under which conditions.

Baseline coverage

Reference methods help show what a model adds beyond simple controls. We track a null control and a conventional method for each protocol.

0 of 2 active baseline roles have published Rewire measurements in this release. Measurements on a selected protocol do not establish coverage of an entire suite.

No execution recipe linked to this protocol. Recipe availability does not establish a completed evaluation.

External evaluations: 5

Literature evidence is not a Rewire measurement. Executed but unpublished runs and private review status are not included.

Null control

Proposed control: requires review

Select a task-valid null control after reviewing inputs and metric

Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.

This is a suggested selection rule, not a validated method or a measured score.

Conventional reference

Proposed control: requires review

Select an upstream conventional reference after reviewing the full protocol

Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.

This is a suggested selection rule, not a validated method or a measured score.

Protocol coverage CSV (gzip) · Model evaluation matrix (gzip) · Source table (gzip) · Release and checksums (gzip)

Coverage is derived from release 2026-10-23-5347695500ed. Source citations describe the original records; they do not validate an unreviewed baseline proposal. No results have been generated by this audit.

Run instructions

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

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.

0 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-10-23-5347695500ed
Property and statement	Original source and location	Review and provenance

No evidence rows match these filters. Choose another scope or clear the search.

How to read the evidence

Sources and history

Release 2026-10-23-5347695500ed · Record review: source checked

1 source record and release history

Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation · Original source · arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19)

Download this release (gzip)

Technical metadata and extraction receipts

Stable ID: lit-cells-networks-20261011-chen2026-protocol-annotation-ataa

areas: single-cell; spatial-omics
contexts: research
protocol: Supervised fine-tuning on reference labels with a stratified 8:2 train/test split; metrics on the shared held-out test set
version: arXiv 2607.17227v2
metric: macro-f1
metric direction: higher
unit: unitless
random seeds: Seed 42 for NumPy, Scanpy and PyTorch (Methods)
limitations: Single run; no spread printed.; No non-foundation-model baseline is reported for this task in the tables.
source locator: Table 4; Methods 'Cell type annotation'

Related records

part of: Harmonised single-cell and spatial foundation model benchmark (Chen et al. 2026)
uses data: ATAA (Chen et al. 2026)
assessment: CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA
assessment: GenePT-w on Chen et al. 2026 supervised cell type annotation, ATAA
assessment: Nicheformer on Chen et al. 2026 supervised cell type annotation, ATAA
assessment: scELMo on Chen et al. 2026 supervised cell type annotation, ATAA
assessment: scGPT-spatial on Chen et al. 2026 supervised cell type annotation, ATAA
assessed by: Transfer cell-type annotations to a new dataset

Suggest a correction