Contested result

scFoundation and scGPT: Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3) reverses the order

Low severity
Not yet reviewed
Ordering reversal
Different tasks across studies

Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3) ranks scGPT above scFoundation in 15 of 16 comparisons; other studies rank scFoundation above scGPT in 4 of 5.

An automatic lead, not a finding of error. The results below are shown as published, everywhere on the site, until a reviewer decides otherwise.

Which studies agree

2 papers rank scFoundation above scGPT in 4 of their 6 comparisons and scGPT higher in 1; 1 is a tie. Csendes et al. 2025 ranks scGPT above scFoundation in 15 of its 20 comparisons and scFoundation higher in 1; 4 are ties.

Rank scFoundation higher (1): Wu et al. 2025
Rank scGPT higher (1): Csendes et al. 2025 (disagrees)
Mixed (1): Deep-learning-based gene perturbation…

Each paper sits where at least two thirds of its comparisons point; Mixed means neither model reaches two thirds. The larger, outlined dot is the paper that disagrees.

scFoundation higherTiescGPT higherNumbers are counts of comparisons; the outlined bar is the paper that disagrees.

Papers compared (3)
Paper	Role	Ranks scFoundation higher	Ranks scGPT higher	Ties
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3) · DOI	Disagrees	1	15	4
Wu et al. 2025, Additional file 3 (Supplementary Tables S1-S13) · DOI	Agrees	3	0	1
Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines and 4 more records · DOI	Agrees	1	1	0

Comparisons in Csendes et al. 2025 (20)

Protocol	Dataset	Metric	scFoundation	scGPT	Higher
Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations (GEARS processing, Csendes et al. 2025)	Pearson (raw expression, pseudo-bulk, all genes), unitless	0.985	0.99	Tie
Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations (GEARS processing, Csendes et al. 2025)	Pearson (raw expression, top 20 DE genes (t-test, GEARS list)), unitless	0.969	0.979	scGPT
Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations (GEARS processing, Csendes et al. 2025)	Pearson delta, unitless	0.552	0.641	scGPT
Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations (GEARS processing, Csendes et al. 2025)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s)), unitless	0.78	0.795	scGPT
Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations (GEARS processing, Csendes et al. 2025)	Pearson delta (top 20 DE genes (t-test, GEARS list)), unitless	0.719	0.797	scGPT
Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations (GEARS processing, Csendes et al. 2025)	Pearson (raw expression, pseudo-bulk, all genes), unitless	0.966	0.988	scGPT
Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations (GEARS processing, Csendes et al. 2025)	Pearson (raw expression, top 20 DE genes (t-test, GEARS list)), unitless	0.924	0.935	scGPT
Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations (GEARS processing, Csendes et al. 2025)	Pearson delta, unitless	0.459	0.554	scGPT
Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations (GEARS processing, Csendes et al. 2025)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s)), unitless	0.71	0.7	scFoundation
Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations (GEARS processing, Csendes et al. 2025)	Pearson delta (top 20 DE genes (t-test, GEARS list)), unitless	0.705	0.732	scGPT
Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi (GEARS processing, Csendes et al. 2025)	Pearson (raw expression, pseudo-bulk, all genes), unitless	0.977	0.986	Tie
Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi (GEARS processing, Csendes et al. 2025)	Pearson (raw expression, top 20 DE genes (t-test, GEARS list)), unitless	0.952	0.96	Tie
Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi (GEARS processing, Csendes et al. 2025)	Pearson delta, unitless	0.269	0.327	scGPT
Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi (GEARS processing, Csendes et al. 2025)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s)), unitless	0.394	0.442	scGPT
Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi (GEARS processing, Csendes et al. 2025)	Pearson delta (top 20 DE genes (t-test, GEARS list)), unitless	0.383	0.481	scGPT
Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi (GEARS processing, Csendes et al. 2025)	Pearson (raw expression, pseudo-bulk, all genes), unitless	0.971	0.974	Tie
Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi (GEARS processing, Csendes et al. 2025)	Pearson (raw expression, top 20 DE genes (t-test, GEARS list)), unitless	0.912	0.924	scGPT
Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi (GEARS processing, Csendes et al. 2025)	Pearson delta, unitless	0.471	0.596	scGPT
Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi (GEARS processing, Csendes et al. 2025)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s)), unitless	0.49	0.629	scGPT
Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi (GEARS processing, Csendes et al. 2025)	Pearson delta (top 20 DE genes (t-test, GEARS list)), unitless	0.495	0.652	scGPT

Comparisons in Wu et al. 2025 (4)

Protocol	Dataset	Metric	scFoundation	scGPT	Higher
Cross-atlas annotation transfer, HLCA to Tabula Sapiens, 14 shared cell types (Wu et al. 2025 Table S3)	Tabula Sapiens (CELLxGENE), 120 leaf cell types after processing	Macro F1 (test cells of the 14 shared cell types; averaged over 'all classes' (Methods 'Macro-F1'), not stated as the 14 shared types or every source-atlas type), fraction	0.259	0.209	scFoundation
Cross-atlas annotation transfer, HLCA to Tabula Sapiens, 14 shared cell types (Wu et al. 2025 Table S3)	Tabula Sapiens (CELLxGENE), 120 leaf cell types after processing	Top-1 accuracy (Accuracy@1 over cells of the 14 shared cell types), fraction	0.782	0.614	scFoundation
Cross-atlas annotation transfer, Tabula Sapiens to HLCA, 14 shared cell types (Wu et al. 2025 Table S2)	HLCA core (CELLxGENE), 37 leaf cell types after processing	Macro F1 (test cells of the 14 shared cell types; averaged over 'all classes' (Methods 'Macro-F1'), not stated as the 14 shared types or every source-atlas type), fraction	0.083	0.081	scFoundation
Cross-atlas annotation transfer, Tabula Sapiens to HLCA, 14 shared cell types (Wu et al. 2025 Table S2)	HLCA core (CELLxGENE), 37 leaf cell types after processing	Top-1 accuracy (Accuracy@1 over cells of the 14 shared cell types), fraction	0.672	0.674	Tie

Comparisons in Deep-learning-based gene perturbation… (2)

Protocol	Dataset	Metric	scFoundation	scGPT	Higher
Norman double perturbations: expression prediction for held-out double perturbations (Ahlmann-Eltze et al. 2025 Fig. 1a and Extended Data Fig. 2a)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations (scFoundation reprocessing, Ahlmann-Eltze et al. 2025)	L2 distance (1,000 most highly expressed control genes; mean over held-out perturbations and splits), unit-unreported	4.72	4.62	scGPT
Norman double perturbations: expression prediction for held-out double perturbations (Ahlmann-Eltze et al. 2025 Fig. 1a and Extended Data Fig. 2a)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations (scFoundation reprocessing, Ahlmann-Eltze et al. 2025)	Pearson delta (1,000 most highly expressed control genes; mean over held-out perturbations and splits), unitless	0.537	0.527	scFoundation

What disagrees

Other studies: rank scFoundation above scGPT in 4 of 5 comparisons
This study: ranks scGPT above scFoundation in 15 of 16 comparisons

Compared: scFoundation, scGPT
Study that disagrees: Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3)
Other studies: 6 studies
Wu et al. 2025, Additional file 3 (Supplementary Tables S1-S13), Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines, Source Data Extended Data Fig. 2, Ahlmann-Eltze et al. 2025, Source Data Extended Data Fig. 8, Ahlmann-Eltze et al. 2025, Source Data Fig. 1, Ahlmann-Eltze et al. 2025, Source Data Fig. 2, Ahlmann-Eltze et al. 2025
Rule: Ordering reversal; different tasks across studies
Model identity: Matched across studies partly through links not yet reviewed

The compared numbers

Each value links to its result record and source. Compare values within one protocol; the rule compares rankings, not raw values across protocols.

In the study that disagrees (32)
Study	Tool	Protocol or benchmark	Metric	Value
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson (raw expression, top 20 DE genes (t-test, GEARS list))	0.969
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta	0.552
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test, GEARS list))	0.719
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s))	0.78
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson (raw expression, top 20 DE genes (t-test, GEARS list))	0.979
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta	0.641
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test, GEARS list))	0.797
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Adamson et al. 2016 Perturb-seq, K562 CRISPRi single perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s))	0.795
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson (raw expression, pseudo-bulk, all genes)	0.966
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson (raw expression, top 20 DE genes (t-test, GEARS list))	0.924
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta	0.459
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test, GEARS list))	0.705
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s))	0.71
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson (raw expression, pseudo-bulk, all genes)	0.988
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson (raw expression, top 20 DE genes (t-test, GEARS list))	0.935
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta	0.554
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test, GEARS list))	0.732
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Norman et al. 2019 Perturb-seq, K562 CRISPRa single and double perturbations: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s))	0.7
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta	0.269
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test, GEARS list))	0.383
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s))	0.394
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta	0.327
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test, GEARS list))	0.481
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, K562 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s))	0.442
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson (raw expression, top 20 DE genes (t-test, GEARS list))	0.912
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta	0.471
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test, GEARS list))	0.495
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scFoundation (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s))	0.49
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson (raw expression, top 20 DE genes (t-test, GEARS list))	0.924
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta	0.596
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test, GEARS list))	0.652
Csendes et al. 2025, Supplementary Material 4 (Supplementary Tables 1-3); Benchmarking foundation cell models for post-perturbation RNA-seq prediction	scGPT (Csendes et al. 2025)	Replogle et al. 2022 genome-wide Perturb-seq subset, RPE1 CRISPRi: unseen-perturbation expression prediction (Csendes et al. 2025 Supplementary Table 2)	Pearson delta (top 20 DE genes (t-test) excluding the CRISPR target gene(s))	0.629

In the other studies (10)
Study	Tool	Protocol or benchmark	Metric	Value
Wu et al. 2025, Additional file 3 (Supplementary Tables S1-S13); Biology-driven insights into the power of single-cell foundation models	scFoundation zero-shot embeddings with OnClass (Wu et al. 2025)	Cross-atlas annotation transfer, HLCA to Tabula Sapiens, 14 shared cell types (Wu et al. 2025 Table S3)	Top-1 accuracy (Accuracy@1 over cells of the 14 shared cell types)	0.782
Wu et al. 2025, Additional file 3 (Supplementary Tables S1-S13); Biology-driven insights into the power of single-cell foundation models	scFoundation zero-shot embeddings with OnClass (Wu et al. 2025)	Cross-atlas annotation transfer, HLCA to Tabula Sapiens, 14 shared cell types (Wu et al. 2025 Table S3)	Macro F1 (test cells of the 14 shared cell types; averaged over 'all classes' (Methods 'Macro-F1'), not stated as the 14 shared types or every source-atlas type)	0.259
Wu et al. 2025, Additional file 3 (Supplementary Tables S1-S13); Biology-driven insights into the power of single-cell foundation models	scGPT zero-shot embeddings with OnClass (Wu et al. 2025)	Cross-atlas annotation transfer, HLCA to Tabula Sapiens, 14 shared cell types (Wu et al. 2025 Table S3)	Top-1 accuracy (Accuracy@1 over cells of the 14 shared cell types)	0.614
Wu et al. 2025, Additional file 3 (Supplementary Tables S1-S13); Biology-driven insights into the power of single-cell foundation models	scGPT zero-shot embeddings with OnClass (Wu et al. 2025)	Cross-atlas annotation transfer, HLCA to Tabula Sapiens, 14 shared cell types (Wu et al. 2025 Table S3)	Macro F1 (test cells of the 14 shared cell types; averaged over 'all classes' (Methods 'Macro-F1'), not stated as the 14 shared types or every source-atlas type)	0.209
Wu et al. 2025, Additional file 3 (Supplementary Tables S1-S13); Biology-driven insights into the power of single-cell foundation models	scFoundation zero-shot embeddings with OnClass (Wu et al. 2025)	Cross-atlas annotation transfer, Tabula Sapiens to HLCA, 14 shared cell types (Wu et al. 2025 Table S2)	Macro F1 (test cells of the 14 shared cell types; averaged over 'all classes' (Methods 'Macro-F1'), not stated as the 14 shared types or every source-atlas type)	0.083
Wu et al. 2025, Additional file 3 (Supplementary Tables S1-S13); Biology-driven insights into the power of single-cell foundation models	scGPT zero-shot embeddings with OnClass (Wu et al. 2025)	Cross-atlas annotation transfer, Tabula Sapiens to HLCA, 14 shared cell types (Wu et al. 2025 Table S2)	Macro F1 (test cells of the 14 shared cell types; averaged over 'all classes' (Methods 'Macro-F1'), not stated as the 14 shared types or every source-atlas type)	0.081
Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines; Source Data Extended Data Fig. 2, Ahlmann-Eltze et al. 2025; Source Data Fig. 1, Ahlmann-Eltze et al. 2025	scFoundation (Ahlmann-Eltze et al. 2025)	Norman double perturbations: expression prediction for held-out double perturbations (Ahlmann-Eltze et al. 2025 Fig. 1a and Extended Data Fig. 2a)	Pearson delta (1,000 most highly expressed control genes; mean over held-out perturbations and splits)	0.537
Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines; Source Data Extended Data Fig. 2, Ahlmann-Eltze et al. 2025; Source Data Fig. 1, Ahlmann-Eltze et al. 2025	scGPT (Ahlmann-Eltze et al. 2025)	Norman double perturbations: expression prediction for held-out double perturbations (Ahlmann-Eltze et al. 2025 Fig. 1a and Extended Data Fig. 2a)	Pearson delta (1,000 most highly expressed control genes; mean over held-out perturbations and splits)	0.527
Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines; Source Data Fig. 1, Ahlmann-Eltze et al. 2025	scFoundation (Ahlmann-Eltze et al. 2025)	Norman double perturbations: expression prediction for held-out double perturbations (Ahlmann-Eltze et al. 2025 Fig. 1a and Extended Data Fig. 2a)	L2 distance (1,000 most highly expressed control genes; mean over held-out perturbations and splits)	4.72
Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines; Source Data Fig. 1, Ahlmann-Eltze et al. 2025	scGPT (Ahlmann-Eltze et al. 2025)	Norman double perturbations: expression prediction for held-out double perturbations (Ahlmann-Eltze et al. 2025 Fig. 1a and Extended Data Fig. 2a)	L2 distance (1,000 most highly expressed control genes; mean over held-out perturbations and splits)	4.62

Follow-up

Current state: Not yet reviewed.

No one has reviewed this finding yet.

How to help resolve this

If you know why these studies disagree, for example a different population, metric or preprocessing step, or you have a corrected value, tell us through the contribution form. A reviewer checks every submission against its source before anything changes, and the step is added to the follow-up above.

Finding discrepancy-a8ccf189542c45d4 · All contested results · How findings are made and followed up