Source

Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation

arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19) · Preprint

DOI: 10.48550/arXiv.2607.17227 Original source

What this source contributed

Records in this release that cite this source. A record can cite several sources.

6 models
21 configurations
1 benchmark
25 protocols
10 dataset subsets
93 evaluations
378 results
7 evidence claims

Results

Values as the source printed them. Compare results only under compatible protocols, datasets and metrics.

Model or configuration	Metric	Value	Dataset	Evaluation
CellPLM (continual, Chen et al. 2026)	Adjusted rand index	0.84 unitless	ATAA (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, ATAA
CellPLM (continual, Chen et al. 2026)	Normalized mutual information	0.82 unitless	ATAA (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, ATAA
CellPLM (continual, Chen et al. 2026)	Silhouette score (against reference labels on the model embedding; range -1 to 1; distance stated as cosine in some model sections of the Methods)	0.26 unitless	ATAA (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, ATAA
CellPLM (continual, Chen et al. 2026)	Adjusted rand index	0.16 unitless	BC (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, BC
CellPLM (continual, Chen et al. 2026)	Normalized mutual information	0.31 unitless	BC (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, BC
CellPLM (continual, Chen et al. 2026)	Silhouette score (against reference labels on the model embedding; range -1 to 1; distance stated as cosine in some model sections of the Methods)	0.10 unitless	BC (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, BC
CellPLM (continual, Chen et al. 2026)	Adjusted rand index	0.28 unitless	BMMC (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, BMMC
CellPLM (continual, Chen et al. 2026)	Normalized mutual information	0.66 unitless	BMMC (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, BMMC
CellPLM (continual, Chen et al. 2026)	Silhouette score (against reference labels on the model embedding; range -1 to 1; distance stated as cosine in some model sections of the Methods)	0.12 unitless	BMMC (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, BMMC
CellPLM (continual, Chen et al. 2026)	Adjusted rand index	0.37 unitless	HS (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, HS
CellPLM (continual, Chen et al. 2026)	Normalized mutual information	0.52 unitless	HS (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, HS
CellPLM (continual, Chen et al. 2026)	Silhouette score (against reference labels on the model embedding; range -1 to 1; distance stated as cosine in some model sections of the Methods)	0.14 unitless	HS (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, HS
CellPLM (continual, Chen et al. 2026)	Adjusted rand index	0.27 unitless	ME (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, ME
CellPLM (continual, Chen et al. 2026)	Normalized mutual information	0.40 unitless	ME (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, ME
CellPLM (continual, Chen et al. 2026)	Silhouette score (against reference labels on the model embedding; range -1 to 1; distance stated as cosine in some model sections of the Methods)	0.09 unitless	ME (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, ME
CellPLM (continual, Chen et al. 2026)	Adjusted rand index	0.15 unitless	MH (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, MH
CellPLM (continual, Chen et al. 2026)	Normalized mutual information	0.21 unitless	MH (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, MH
CellPLM (continual, Chen et al. 2026)	Silhouette score (against reference labels on the model embedding; range -1 to 1; distance stated as cosine in some model sections of the Methods)	0.06 unitless	MH (Chen et al. 2026)	CellPLM on Chen et al. 2026 continually pretrained clustering, MH
CellPLM (gears-decoder, Chen et al. 2026)	MSE (all genes, predicted versus observed post-perturbation expression)	0.0091 unitless	Adamson (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Adamson
CellPLM (gears-decoder, Chen et al. 2026)	MSE (top 20 differentially expressed genes per perturbation condition, predicted versus observed post-perturbation expression)	0.2778 unitless	Adamson (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Adamson
CellPLM (gears-decoder, Chen et al. 2026)	Pearson (all genes, predicted versus observed post-perturbation expression)	0.9876 unitless	Adamson (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Adamson
CellPLM (gears-decoder, Chen et al. 2026)	Pearson (top 20 differentially expressed genes per perturbation condition, predicted versus observed post-perturbation expression)	0.9236 unitless	Adamson (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Adamson
CellPLM (gears-decoder, Chen et al. 2026)	Pearson delta (all genes in the intersected gene set)	0.6487 unitless	Adamson (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Adamson
CellPLM (gears-decoder, Chen et al. 2026)	Pearson delta (top 20 differentially expressed genes per perturbation condition)	0.7112 unitless	Adamson (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Adamson
CellPLM (gears-decoder, Chen et al. 2026)	MSE (all genes, predicted versus observed post-perturbation expression)	0.0064 unitless	Norman (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Norman
CellPLM (gears-decoder, Chen et al. 2026)	MSE (top 20 differentially expressed genes per perturbation condition, predicted versus observed post-perturbation expression)	0.4034 unitless	Norman (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Norman
CellPLM (gears-decoder, Chen et al. 2026)	Pearson (all genes, predicted versus observed post-perturbation expression)	0.9826 unitless	Norman (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Norman
CellPLM (gears-decoder, Chen et al. 2026)	Pearson (top 20 differentially expressed genes per perturbation condition, predicted versus observed post-perturbation expression)	0.8724 unitless	Norman (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Norman
CellPLM (gears-decoder, Chen et al. 2026)	Pearson delta (all genes in the intersected gene set)	0.4727 unitless	Norman (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Norman
CellPLM (gears-decoder, Chen et al. 2026)	Pearson delta (top 20 differentially expressed genes per perturbation condition)	0.3982 unitless	Norman (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Norman
CellPLM (gears-decoder, Chen et al. 2026)	MSE (all genes, predicted versus observed post-perturbation expression)	0.0068 unitless	Replogle (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Replogle
CellPLM (gears-decoder, Chen et al. 2026)	MSE (top 20 differentially expressed genes per perturbation condition, predicted versus observed post-perturbation expression)	0.1844 unitless	Replogle (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Replogle
CellPLM (gears-decoder, Chen et al. 2026)	Pearson (all genes, predicted versus observed post-perturbation expression)	0.9868 unitless	Replogle (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Replogle
CellPLM (gears-decoder, Chen et al. 2026)	Pearson (top 20 differentially expressed genes per perturbation condition, predicted versus observed post-perturbation expression)	0.9406 unitless	Replogle (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Replogle
CellPLM (gears-decoder, Chen et al. 2026)	Pearson delta (all genes in the intersected gene set)	0.3801 unitless	Replogle (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Replogle
CellPLM (gears-decoder, Chen et al. 2026)	Pearson delta (top 20 differentially expressed genes per perturbation condition)	0.4604 unitless	Replogle (Chen et al. 2026)	CellPLM on Chen et al. 2026 genetic perturbation prediction, Replogle
CellPLM (fine-tuned, Chen et al. 2026)	Accuracy (held-out test set)	0.97 unitless	ATAA (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA
CellPLM (fine-tuned, Chen et al. 2026)	AUPRC (held-out test set; macro-averaged over classes)	0.98 unitless	ATAA (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA
CellPLM (fine-tuned, Chen et al. 2026)	AUROC (held-out test set; macro-averaged over classes)	0.99 unitless	ATAA (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA
CellPLM (fine-tuned, Chen et al. 2026)	Macro F1 (held-out test set)	0.95 unitless	ATAA (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA
CellPLM (fine-tuned, Chen et al. 2026)	Precision (held-out test set; described as global in the Methods, averaging not further defined)	0.97 unitless	ATAA (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA
CellPLM (fine-tuned, Chen et al. 2026)	Recall (held-out test set; described as global in the Methods, averaging not further defined)	0.97 unitless	ATAA (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, ATAA
CellPLM (fine-tuned, Chen et al. 2026)	Accuracy (held-out test set)	0.86 unitless	BMMC (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, BMMC
CellPLM (fine-tuned, Chen et al. 2026)	AUPRC (held-out test set; macro-averaged over classes)	0.80 unitless	BMMC (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, BMMC
CellPLM (fine-tuned, Chen et al. 2026)	AUROC (held-out test set; macro-averaged over classes)	0.99 unitless	BMMC (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, BMMC
CellPLM (fine-tuned, Chen et al. 2026)	Macro F1 (held-out test set)	0.73 unitless	BMMC (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, BMMC
CellPLM (fine-tuned, Chen et al. 2026)	Precision (held-out test set; described as global in the Methods, averaging not further defined)	0.86 unitless	BMMC (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, BMMC
CellPLM (fine-tuned, Chen et al. 2026)	Recall (held-out test set; described as global in the Methods, averaging not further defined)	0.86 unitless	BMMC (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, BMMC
CellPLM (fine-tuned, Chen et al. 2026)	Accuracy (held-out test set)	0.93 unitless	HS (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, HS
CellPLM (fine-tuned, Chen et al. 2026)	AUPRC (held-out test set; macro-averaged over classes)	0.96 unitless	HS (Chen et al. 2026)	CellPLM on Chen et al. 2026 supervised cell type annotation, HS

Claims

Statements this source supports, each with where in the source it was checked.

Relevance of Adamson K562 Perturb-seq to "Assess models for genetic perturbation experiments": Assessed by Chen et al. 2026 genetic perturbation prediction, Adamson (lit-cells-networks-20261011-chen2026-source-chen2026: Table 5; Results 'Perturbation prediction separates expression reconstruction from functional response modelling')
Relevance of ATAA held-out annotation to "Transfer cell-type annotations to a new dataset": Assessed by Chen et al. 2026 supervised cell type annotation, ATAA (lit-cells-networks-20261011-chen2026-source-chen2026: Table 4; Results on supervised annotation; Methods 'Cell type annotation')
Relevance of BMMC held-out annotation to "Transfer cell-type annotations to a new dataset": Assessed by Chen et al. 2026 supervised cell type annotation, BMMC (lit-cells-networks-20261011-chen2026-source-chen2026: Table 4; Results on supervised annotation; Methods 'Cell type annotation')
Relevance of HS held-out annotation to "Transfer cell-type annotations to a new dataset": Assessed by Chen et al. 2026 supervised cell type annotation, HS (lit-cells-networks-20261011-chen2026-source-chen2026: Table 4; Results on supervised annotation; Methods 'Cell type annotation')
Relevance of Norman K562 Perturb-seq to "Assess models for genetic perturbation experiments": Assessed by Chen et al. 2026 genetic perturbation prediction, Norman (lit-cells-networks-20261011-chen2026-source-chen2026: Table 5; Results 'Perturbation prediction separates expression reconstruction from functional response modelling')
Relevance of Replogle K562 Perturb-seq to "Assess models for genetic perturbation experiments": Assessed by Chen et al. 2026 genetic perturbation prediction, Replogle (lit-cells-networks-20261011-chen2026-source-chen2026: Table 5; Results 'Perturbation prediction separates expression reconstruction from functional response modelling')
Relevance of SEAAD held-out annotation to "Transfer cell-type annotations to a new dataset": Assessed by Chen et al. 2026 supervised cell type annotation, SEAAD (lit-cells-networks-20261011-chen2026-source-chen2026: Table 4; Results on supervised annotation; Methods 'Cell type annotation')

Models, benchmarks and other records

Models (6)

Configurations (21)

Benchmarks (1)

Harmonised single-cell and spatial foundation model benchmark (Chen et al. 2026)

Protocols (25)

Dataset subsets (10)

Evaluations (93)

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.

24 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
Artifact SHA-256 8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da Source metadata	Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation Original source ↗ No field-specific location recorded Version: arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19) Retrieved: 2026-10-11T21:29:34Z	catalogued No individual claim review recorded Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.artifact_sha256` Source artifact SHA-256: `8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Artifact url https://arxiv.org/pdf/2607.17227v2 Source metadata	Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation Original source ↗ No field-specific location recorded Version: arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19) Retrieved: 2026-10-11T21:29:34Z	catalogued No individual claim review recorded Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.artifact_url` Source artifact SHA-256: `8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Doi 10.48550/arXiv.2607.17227 Source metadata	Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation Original source ↗ No field-specific location recorded Version: arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19) Retrieved: 2026-10-11T21:29:34Z	catalogued No individual claim review recorded Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.doi` Source artifact SHA-256: `8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Evidence concerns 1 values source id lit-cells-networks-20261011-chen2026-source-chen2026 message 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. source locator Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25) artifact sha256 8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da review date 2026-10-12 Source metadata	Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation Original source ↗ No field-specific location recorded Version: arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19) Retrieved: 2026-10-11T21:29:34Z	catalogued No individual claim review recorded Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.evidence_concerns` Source artifact SHA-256: `8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Licence CC-BY-SA-4.0 Source metadata	Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation Original source ↗ No field-specific location recorded Version: arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19) Retrieved: 2026-10-11T21:29:34Z	catalogued No individual claim review recorded Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.licence` Source artifact SHA-256: `8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Limitations 3 values Preprint, not peer reviewed. Single run per cell with a fixed seed (42); no spread is printed. Spatial dataset codes BC, MH and ME are not defined in the tables; the Figure 2 legend defines them as BC/HBCA1 (human breast cancer), MH/MHPC (mouse hippocampus) and ME/MOSTA E9.5 (mouse embryo). Source metadata	Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation Original source ↗ No field-specific location recorded Version: arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19) Retrieved: 2026-10-11T21:29:34Z	catalogued No individual claim review recorded Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.limitations` Source artifact SHA-256: `8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Media type application/pdf Source metadata	Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation Original source ↗ No field-specific location recorded Version: arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19) Retrieved: 2026-10-11T21:29:34Z	catalogued No individual claim review recorded Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.media_type` Source artifact SHA-256: `8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Publication status preprint Source metadata	Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation Original source ↗ No field-specific location recorded Version: arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19) Retrieved: 2026-10-11T21:29:34Z	catalogued No individual claim review recorded Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.publication_status` Source artifact SHA-256: `8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Retrieved at 2026-10-11T21:29:34Z Source metadata	Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation Original source ↗ No field-specific location recorded Version: arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19) Retrieved: 2026-10-11T21:29:34Z	catalogued No individual claim review recorded Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.retrieved_at` Source artifact SHA-256: `8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact
Review: date 2026-10-12 Source metadata	Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation Original source ↗ No field-specific location recorded Version: arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19) Retrieved: 2026-10-11T21:29:34Z	catalogued No individual claim review recorded Source has a recorded evidence concern. Consult its source page before using the claim. Audit details Field: `attributes.review.date` Source artifact SHA-256: `8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da` Hash scope: Hash scope not separately documented; inspect source record Inspected artifact

How to read the evidence

Sources and history

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

Primary source retrieved and hashed for the 2026-10-11 cells-spatial and networks-mechanistic literature pass.

Release history

No supporting source is linked yet.

Read original source

Download this release (gzip)

Technical metadata and extraction receipts

Stable ID: lit-cells-networks-20261011-chen2026-source-chen2026

areas: single-cell; spatial-omics
contexts: research
url: https://arxiv.org/abs/2607.17227v2
artifact url: https://arxiv.org/pdf/2607.17227v2
version: arXiv 2607.17227v2, updated 2026-09-17 (v1 posted 2026-07-19)
retrieved at: 2026-10-11T21:29:34Z
artifact sha256: 8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da
doi: 10.48550/arXiv.2607.17227
publication status: preprint
licence: CC-BY-SA-4.0
media type: application/pdf
limitations: Preprint, not peer reviewed.; Single run per cell with a fixed seed (42); no spread is printed.; Spatial dataset codes BC, MH and ME are not defined in the tables; the Figure 2 legend defines them as BC/HBCA1 (human breast cancer), MH/MHPC (mouse hippocampus) and ME/MOSTA E9.5 (mouse embryo).
evidence concerns: source id: lit-cells-networks-20261011-chen2026-source-chen2026; message: 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.; source locator: Table 4, columns 'Accuracy', 'Precision' and 'Recall'; Methods 'Cell type annotation' (page 25); artifact sha256: 8137338c6146f570a6546f30707d4244baca9dab848b32c51539665a5ba539da; review date: 2026-10-12
review: method: source-hash-verification; method note: Artifact re-downloaded from arXiv and SHA-256 matched; the archived gzip copy decompresses to the same bytes.; reviewer: claude; reviewer note: Claude review agent reviewer-cn-2, independent of the extractor lit-cells-networks; no human review claimed; date: 2026-10-12

Related records

Suggest a correction