A systematic comparison of single-cell perturbation response prediction models
Open accessThe DOI matches its Crossref record; 69 of the 69 references it cites were found in a registry or an index; 27 of the 33 sources its claims cite were checked for support.
Abstract
Predicting single-cell transcriptional responses to perturbations is central to dissecting gene regulation and accelerating therapeutic design, yet the field lacks a rigorous, task-spanning assessment of model behavior. We present a large-scale benchmark of 13 representative methods and baselines across 25 datasets spanning diverse perturbation modalities and species, including two primary immune-cell drug-response resources. We evaluated three core tasks-generalization to unseen single-gene perturbations, prediction of combinatorial interactions, and transfer across cell types-using 24 metrics covering expression-level accuracy, relative changes, differential expression (DE) recovery, and distributional similarity. Across tasks, performance depended strongly on perturbation effect size and evaluation perspective: Expression-level agreement was the highest for small-effect perturbations resembling controls, whereas delta- and DE-based metrics improved with larger effects, providing clearer signals. Models shared a conservative bias, with fine-tuned foundation models compressing variance and underestimating synergistic effects in combinations. PerturbNet showed superior recovery of DE signatures in Tasks 1 and 2, while no method consistently generalized across cell types in Task 3, where biological consistency dominated outcomes. This benchmark establishes current methodological limits, clarifies that different metrics probe distinct biological signals rather than redundant summaries of the same prediction problem, and provides a foundation for developing virtual-cell models that more faithfully capture heterogeneous perturbation responses.
Overview
CoChat Summary
TL;DR
This paper presents a large-scale benchmark of 13 single-cell perturbation response prediction methods and baselines across 25 datasets, three prediction tasks, and 24 metrics. It finds that performance depends strongly on perturbation effect size and evaluation metric, with no method consistently generalizing across cell types and with fine-tuned foundation models showing variance compression.
Key findings
- Across 25 datasets and 24 metrics, absolute expression-level agreement was often high, with Pearson’s correlation exceeding 0.95 for most datasets, but this did not consistently translate into accurate recovery of perturbation-induced changes, DEGs, distributional similarity, or cross-cell-type transfer.
- PerturbNet showed superior recovery of DE signatures in Tasks 1 and 2, achieving a mean overlap accuracy of 0.56 in Task 1, while most other methods failed to recover DEGs reliably in several datasets.
- Performance depended strongly on perturbation effect size: small-effect perturbations resembling controls yielded higher expression-level agreement, whereas delta- and DE-based metrics improved with larger effects.
- In Task 2, simple baselines such as additive mean and linear models were highly competitive and often surpassed complex architectures in seen1 and seen2 combinatorial settings. In Task 3, no method consistently generalized across cell types, and biological consistency dominated outcomes.
Methods
- Benchmark design: 13 representative methods and baselines were evaluated across three tasks—unseen single-gene perturbation prediction, combinatorial perturbation interaction prediction, and cell state transfer—using 25 datasets spanning genetic perturbations, small-molecule treatments, immune stimuli, cancer cell lines, pluripotent stem cells, primary immune populations, and nonhuman immune cells.
- Data and models: The collection integrated 23 public datasets and 2 in-house datasets, including two primary immune-cell drug-response resources; models included foundation models (scGPT, scFoundation), GRN-based models (GEARS, scELMo), VAE-based models (PerturbNet, scGen, trVAE, scVIDR, scPRAM), and other deep learning models (CPA, scPreGAN, BioLord, Squidiff), plus task-specific statistical and model-based baselines.
- Analysis: Evaluation used 24 metrics covering absolute accuracy, relative effect capture, DE recovery, and distributional similarity, with ranking-based meta-analysis and ANOVA variance decomposition to assess contributions of model choice, dataset identity, perturbation identity, effect size, and other factors.
Limitations
- The paper states that datasets were generated using different technologies, perturbation modalities, cell types, and experimental protocols, introducing batch effects and data-quality variation that can influence training and evaluation; variance analysis indicated intrinsic dataset properties can contribute as much to performance variation as model choice.
- The paper states that current single-cell sequencing is destructive, so the same cell cannot be observed before and after perturbation; all models infer population-level counterfactuals rather than true paired single-cell trajectories, and the lack of dense temporal sampling limits distinguishing direct perturbation effects from downstream adaptation.
- The text makes evident that PerturbNet requires chemical structure information (SMILES) and is therefore only applicable to drug perturbation datasets, so not all models were evaluated on all dataset types; Task 3 evaluated only six methods, and DE recovery remained poor across all models in that task.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
11 Valid · 2 Not valid · 0 Not checked · 81 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“Besides, simple baselines proved highly competitive with complex deep learning models, which is consistent with a previous study (25).”
§ A systematic comparison of single-cell perturbation response prediction models > RESULTS > Perturbation interaction prediction
What the sources say
- Supportedcites [25] C. Ahlmann-Eltze, W. Huber, S. Anders, Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines. Nat. Methods 22, 1657– 1661 (2025). 40759747 PMC12328236 source
The paper’s summary of its benchmark states that current foundation models did not perform better than deliberately simplistic linear prediction models. Its double-perturbation results also report substantially higher prediction error for the tested models than for the additive baseline. These passages support the citing sentence’s broad claim that simple baselines were highly competitive with complex deep-learning models. (LLM, conf=0.94)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Compared to tasks such as cell annotation and batch correction, perturbation prediction offers experimentally generated model input/output with well-defined objectives, making it a more robust benchmark for assessing foundation models (15, 20, 21).”
§ A systematic comparison of single-cell perturbation response prediction models > INTRODUCTION
What the sources say
- Partly supportedcites [15] H. Cui, C. Wang, H. Maan, K. Pang, F. Luo, N. Duan, B. Wang, scGPT: Toward building a foundation model for single-cell multi-omics using generative AI. Nat. Methods 21, 1470– 1480 (2024). 38409223 sourceExperimental — accuracy not yet measured
Methods §4.5 describes control cells as inputs and experimentally perturbed cells as targets; §4.7 specifies metrics for evaluating predictions against measured expression. These passages support the claim that perturbation prediction has defined inputs, outputs and objectives, but the paper does not say it is a more robust foundation-model benchmark than cell annotation or batch correction. Page numbers were not supplied for these passages. (LLM, conf=0.83)
- Partly supportedcites [20] X. Shao, H. Yang, X. Zhuang, J. Liao, P. Yang, J. Cheng, X. Lu, H. Chen, X. Fan, scDeepSort: A pre-trained cell-type annotation method for single-cell transcriptomics using deep learning with a weighted graph neural network. Nucleic Acids Res. 49, e122 (2021). 34500471 PMC8643674 sourceExperimental — accuracy not yet measured
The Results section identifies scDeepSort as a cell-type annotation method and reports its performance on annotation datasets, supporting cell annotation as an example of a task. The paper does not address perturbation prediction, experimentally generated input/output, or whether that task is a more robust benchmark for foundation models than annotation and batch correction (searched those terms). The indexed passages provide no page numbers. (LLM, conf=0.86)
- Partly supportedcites [21] M. Hao, J. Gong, X. Zeng, C. Liu, Y. Guo, X. Cheng, T. Wang, J. Ma, X. Zhang, L. Song, Large-scale foundation model on single-cell transcriptomics. Nat. Methods 21, 1481– 1491 (2024). 38844628 sourceExperimental — accuracy not yet measured
In “scFoundation predicts more accurate perturbation responses” (page unavailable), the paper describes experimentally measured perturbation responses, pre- and post-perturbation expression profiles, and defined evaluation measures such as mean squared error against ground truth. It does not say that perturbation prediction is a more robust benchmark than cell annotation or batch correction, or make that comparison. (LLM, conf=0.85)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“This finding extends recent observations that simple additive or linear baselines remain difficult to outperform in double-perturbation prediction (25, 31).”
§ A systematic comparison of single-cell perturbation response prediction models > DISCUSSION
What the sources say
- Supportedcites [25] C. Ahlmann-Eltze, W. Huber, S. Anders, Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines. Nat. Methods 22, 1657– 1661 (2025). 40759747 PMC12328236 source
The Main section reports that every evaluated model had substantially higher prediction error than the additive baseline on held-out double perturbations. The Methods section identifies the additive model as a special case of the linear model, supporting the paraphrase about simple additive or linear baselines. (LLM, conf=0.93)
- Not checked — no full text of the sourcecites [31] Z. Wei, Y. Wang, Y. Gao, S. Wang, P. Li, D. Si, Y. Gao, S. Wu, D. Li, K. Dong, X. Yang, C. Tang, S. Fu, X. Chen, W. Li, Y. You, C. Zhang, A. Liang, G. Chuai, Q. Liu, Benchmarking algorithms for generalizable single-cell perturbation response prediction. Nat. Methods 23, 451– 464 (2026). 41381899 source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Recent advances have incorporated graph-based methods and foundational models that take advantage of prior knowledge of gene expression relationships, resulting in improved generalizability (14–17).”
§ A systematic comparison of single-cell perturbation response prediction models > INTRODUCTION
What the sources say
- Partly supportedcites [14] Y. Roohani, K. Huang, J. Leskovec, Predicting transcriptional outcomes of novel multigene perturbations with GEARS. Nat. Biotechnol. 42, 927– 935 (2024). 37592036 PMC11180609 sourceExperimental — accuracy not yet measured
The Main and Results sections describe GEARS as a graph-based method that uses prior knowledge of gene relationships and improves prediction for previously unperturbed genes. The cited paper does not identify GEARS as a foundation model or establish the sentence’s claim about foundation models; that portion is unsupported by this source. (LLM, conf=0.85)
- Partly supportedcites [15] H. Cui, C. Wang, H. Maan, K. Pang, F. Luo, N. Duan, B. Wang, scGPT: Toward building a foundation model for single-cell multi-omics using generative AI. Nat. Methods 21, 1470– 1480 (2024). 38409223 sourceExperimental — accuracy not yet measured
The Discussion (p. 17) says scGPT is a pretrained foundation model that generalizes, captures gene networks, and improves downstream performance through transfer learning. Results §2.5 (p. 12) says it learns gene–gene relationships from expression data, but does not describe incorporating a graph-based method or supplying those relationships as prior knowledge. This source supports the foundation-model and generalizability parts, not that more specific account of how the improvement was achieved. (LLM, conf=0.83)
- Not checked — no full text of the sourcecites [16] J. E. Rood, A. Hupalowska, A. Regev, Toward a foundation model of causal cell and tissue biology with a Perturbation Cell and Tissue Atlas. Cell 187, 4520– 4545 (2024). 39178831 source
- Partly supportedcites [17] G. I. Gavriilidis, V. Vasileiou, A. Orfanou, N. Ishaque, F. Psomopoulos, A mini-review on perturbation modelling across single-cell omic modalities. Comput. Struct. Biotechnol. J. 23, 1886– 1896 (2024). 38721585 PMC11076269 sourceExperimental — accuracy not yet measured
The Complex generative models section describes a graph-based method that uses gene co-expression and other biological priors to predict unseen perturbation responses. The Foundational models section says these models can extrapolate from known experiments to unknown scenarios, but frames this as an aspiration; the Summary and outlook also cautions about weaknesses in zero-shot learning. The review therefore supports the methods and their potential to generalize, not the unqualified claim that they have improved generalizability. (LLM, conf=0.84)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“However, these methods could not be generalized to unseen perturbations.”
§ A systematic comparison of single-cell perturbation response prediction models > INTRODUCTION
What the sources say
- Could not verifycites [11] M. W. Fiers, L. Minnoye, S. Aibar, C. Bravo González-Blas, Z. Kalender Atak, S. Aerts, Mapping gene regulatory networks from single-cell omics data. Brief. Funct. Genomics 17, 246– 254 (2018). 29342231 PMC6063279 sourceExperimental — accuracy not yet measured
The ‘Single-cell regulatory perturbations’ section describes methods for measuring and analysing applied perturbations, but does not say whether they generalize to unseen ones. A search for generalization to new perturbations found no passage addressing that limitation. (LLM, conf=0.72)
- Could not verifycites [12] R. Lopez, A. Gayoso, N. Yosef, Enhancing scientific discoveries in molecular biology with deep generative models. Mol. Syst. Biol. 16, e9198 (2020). 32975352 PMC7517326 sourceExperimental — accuracy not yet measured
The 'In silico generation of biomedical data' section describes scGen predicting a perturbation’s effect in a cell type lacking perturbation data. The 'Disentangling factors of variation' section questions whether its assumptions hold for cell-type-specific responses. Neither passage says that early methods could not generalize to unseen perturbations; searches for unseen and novel perturbations found no such statement. Page numbers were not provided in our copy. (LLM, conf=0.70)
- Not checked — no full text of the sourcecites [13] Y. Ji, M. Lotfollahi, F. A. Wolf, F. J. Theis, Machine learning for perturbational single-cell omics. Cell Syst. 12, 522– 537 (2021). 34139164 source
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
48 more claims · 1 with their cited sources checked
Create a free account to see all 94 claims and every source verdict.
- It may state two separate claims as one. 33
- Probably not a substantial claim. 12
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ A systematic comparison of single-cell perturbation response prediction models > RESULTS > Prediction of unseen single-gene perturbations
“In this setting, PerturbNet emerged as the clear top performer, achieving a mean overlap accuracy of 0.56”
§ A systematic comparison of single-cell perturbation response prediction models > MATERIALS AND METHODS > Model selection and setup
“Note that PerturbNet requires chemical structure information (SMILES representations) as input and is therefore only applicable to drug perturbation datasets.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ A systematic comparison of single-cell perturbation response prediction models > MATERIALS AND METHODS > Evaluation metrics
“This metric takes values of 0 or 1, where 1 indicates perfect directional agreement and 0 indicates directional disagreement.”
§ A systematic comparison of single-cell perturbation response prediction models > MATERIALS AND METHODS > Evaluation metrics
“The delta direction accuracy is then defined as the proportion of genes where the predicted and observed expression changes have the same directional sign, calculated asΔ Agreement acc.=1n∑i=1n I[sign(Δyˆi)=sign(Δyi)](17)”
Citations
Cites
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173
Dixit, A., Parnas, O., Li, B., Chen, J., Fulco, C. P., Jerby-Arnon, L., Marjanovic, N. D., Dionne, D., Burks, T., Raychowdhury, R., et al. (2016) · Cell · doi:10.1016/j.cell.2016.11.038 · Cited by 1,863
Self-citationA single-cell survey of the small intestinal epithelium
Haber, A. L., Biton, M., Rogel, N., Herbst, R. H., Shekhar, K., Smillie, C., Burgin, G., Delorey, T. M., Howitt, M. R., Katz, Y., et al. (2017) · Nature · doi:10.1038/nature24489 · Cited by 1,750
scGPT: Towards Building a Foundation Model for Single-Cell Multi-omics Using Generative AI
Cui, H., Wang, C., Maan, H., Pang, K., Luo, F., & Wang, B. (2023) · doi:10.1101/2023.04.30.538439 · Cited by 1,694
The support of human genetic evidence for approved drug indications
Nelson, M. R., Tipney, H., Painter, J. L., Shen, J., Nicoletti, P., Shen, Y., Floratos, A., Sham, P. C., Li, M. J., Wang, J., et al. (2015) · Nature Genetics · doi:10.1038/ng.3314 · Cited by 1,666
Adamson, B., Norman, T. M., Jost, M., Cho, M. Y., Nuñez, J. K., Chen, Y., Villalta, J. E., Gilbert, L. A., Horlbeck, M. A., Hein, M. Y., et al. (2016) · Cell · doi:10.1016/j.cell.2016.11.048 · Cited by 1,193
Transfer learning enables predictions in network biology
Theodoris, C. V., Xiao, L., Chopra, A., Chaffin, M. D., Al Sayed, Z. R., Hill, M. C., Mantineo, H., Brydon, E. M., Zeng, Z., Liu, X. S., et al. (2023) · Nature · doi:10.1038/s41586-023-06139-9 · Cited by 1,193
Self-citationMultiplexed droplet single-cell RNA-sequencing using natural genetic variation
Kang, H. M., Subramaniam, M., Targ, S., Nguyen, M., Maliskova, L., McCarthy, E., Wan, E., Wong, S., Byrnes, L., Lanata, C. M., et al. (2018) · Nature Biotechnology · doi:10.1038/nbt.4042 · Cited by 1,101
Pooled CRISPR screening with single-cell transcriptome readout.
Datlinger, P., Rendeiro, A. F., Schmidl, C., Krausgruber, T., Traxler, P., Klughammer, J., Schuster, L. C., Kuchler, A., Alpar, D., & Bock, C. (2017) · Nature Methods · doi:10.1038/nmeth.4177 · Cited by 1,096
Mapping information-rich genotype-phenotype landscapes with genome-scale Perturb-seq
Replogle, J. M., Saunders, R. A., Pogson, A. N., Hussmann, J. A., Lenail, A., Guna, A., Mascibroda, L., Wagner, E. J., Adelman, K., Lithwick-Yanai, G., et al. (2022) · Cell · doi:10.1016/j.cell.2022.05.013 · Cited by 795
Cited by
Ahlmann-Eltze, C., Huber, W., & Anders, S. (2025) · Nature Methods · doi:10.1038/s41592-025-02772-6 · Cited by 238
Predicting cellular responses to perturbation across diverse contexts with State
Adduri, A. K., Gautam, D., Bevilacqua, B., Naghipourfar, M., Imran, A., Shah, R., Teyssier, N., Verma, R., Carpenter, C., Eraslan, B., et al. (2026) · Cell · doi:10.1016/j.cell.2026.07.052 · Cited by 100
Huang, A. C., Hsieh, T.-H. S., Zhu, J., Michuda, J., Teng, A., Kim, S., Rumsey, E. M., Lam, S. K., Anigbogu, I., Wright, P., et al. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.06.11.659105 · Cited by 68
A comparison of computational methods for expression forecasting
Kernfeld, E. M., Yang, Y., Weinstock, J. S., Battle, A. J., & Cahan, P. (2025) · Genome biology · doi:10.1186/s13059-025-03840-y · Cited by 50
Yiu, T., Chen, B., Wang, H., Feng, G., Fu, Q.-Q., & Hu, H. (2025) · Journal of Translational Medicine · doi:10.1186/s12967-025-07091-0 · Cited by 40
Wong, D. R., Hill, A. S., & Moccia, R. (2025) · Bioinformatics · doi:10.1093/bioinformatics/btaf317 · Cited by 27
Guo, Y., Zhang, H., Hu, H., Wu, J., Cao, J., Hsieh, C., & Yang, B. (2026) · Nature Machine Intelligence · doi:10.1038/s42256-025-01165-w · Cited by 25
Virtual Cells Need Context, Not Just ScalePreprint
Dibaeinia, P., Babu, S., Knudson, M. K., ElSheikh, A., Wen, Y., Liu, H., Perera, J. F., & Khan, A. A. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.02.04.703804 · Cited by 22
Machine learning to dissect perturbations in complex cellular systems
Monfort‐Lanzas, P., Rungger, K., Madersbacher, L., & Hackl, H. (2025) · Computational and Structural Biotechnology Journal · doi:10.1016/j.csbj.2025.02.028 · Cited by 23
Deep learning perturbation models can outperform baselines on calibrated metrics
Miller, H. E., Mejia, G. M., Leblanc, F. J. A., Swain, B., Wang, B., & de Lima Camillo, L. P. (2026) · Nature Biotechnology · doi:10.1038/s41587-026-03307-w · Cited by 18
GPerturb: Gaussian process modelling of single-cell perturbation data
Xing, H., & Yau, C. (2025) · Nature Communications · doi:10.1038/s41467-025-61165-7 · Cited by 14
scArchon: a scalable benchmarking framework for assessing single-cell perturbation models
Radig, J., Droit, R., Doncevic, D., Li, A., Bui, D., Herfurth, L., Kühn, T., & Herrmann, C. (2026) · Genome biology · doi:10.1186/s13059-026-04104-z · Cited by 13
Mapping transcriptional responses to cellular perturbations with RNA fingerprinting
Grabski, I. N., Lee, J., Blair, J. D., Dalgarno, C., Mascio, I., Bradu, A., Knowles, D. A., & Satija, R. (2026) · Cell · doi:10.1016/j.cell.2026.09.017 · Cited by 8
Hasanaj, E., Cole, E., Mohammadi, S., Addagudi, S., Zhang, X., Song, L., & Xing, E. P. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.06.26.661186 · Cited by 7
Benchmarking biomedical foundation models
Sáez-Rodríguez, J., Schäfer, P. S. L., Kalavros, N., & Stolovitzky, G. A. (2026) · Nature Methods · doi:10.1038/s41592-026-03182-y · Cited by 7
Latent causal diffusions for single-cell perturbation modeling
Lorch, L., Zhang, J., Bunne, C., Krause, A., Schölkopf, B., & Uhler, C. (2026) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2602630123 · Cited by 6
Consequences of training data composition for deep learning models in single-cell biologyPreprint
Nadig, A., Thoutam, A., Hughes, M., Gupta, A., Navia, A. W., Fusi, N., Raghavan, S., Winter, P. S., Amini, A. P., & Crawford, L. A. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.02.19.639127 · Cited by 5
Evaluating Single-Cell Perturbation Response Models Is Far from StraightforwardPreprint
Heidari, M., Karimpour, M., Srivatsa, S., & Montazeri, H. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.02.14.705879 · Cited by 7
Towards Principled Evaluation of Single-Cell Perturbation Prediction ModelsPreprint
Schäfer, P. S. L., Reid, K. A., Boldyga, Z., Aksu, E. D., Hakem, H., & Sáez-Rodríguez, J. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.07.23.740433 · Cited by 6
Unify learns cellular evolution with universal multimodal embeddings
Zhong, H., Han, W., Cui, G., Gómez-Cabrero, D., Tegner, J. N., Gao, X., & Aranda, M. (2026) · Nature Communications · doi:10.1038/s41467-026-76230-y · Cited by 2
A knowledge-driven framework for predicting single-cell responses for unprofiled drugs
Feng, J., Zhao, Z., Zhang, X., Liu, M., Chen, J., Quan, X., Fu, B., Zhang, J., Wang, Y., Zhang, Y., et al. (2026) · Nature Machine Intelligence · doi:10.1038/s42256-026-01286-w · Cited by 6
Self-citationYin, D., Zhang, Z., Liu, X., Ni, K., Su, H., Li, N. L., Dong, H., Zhao, Q., Lin, X., Tian, L., et al. (2025) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2025.07.15.663271 · Cited by 1
Self-citationShen, H. (2026) · Research Square · doi:10.21203/rs.3.rs-10846736/v1 · Cited by 1
Charting the computational landscape of single-cell genetic perturbation
Zhang, Y., Huang, J., Li, Y., Liu, T., Ren, L., Huang, J. H., Lin, H., Feng, J., & Zhang, Y. (2026) · Journal of Advanced Research · doi:10.1016/j.jare.2026.06.012 · Cited by 1
PertDiffBench: Benchmarking Diffusion Models for Single-Cell Perturbation Response PredictionPreprint
Song, Z., Xiang, Y., Song, Z., Jin, W. T., Li, J., Sun, C., & Xie, L. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.06.13.732013
Sun, B., Zheng, H., Chen, J., & Jin, J. (2026) · Genes · doi:10.3390/genes17070816
Self-citationA benchmarking framework for perturbational signature recontextualization across biological contextsPreprint
Chen, A., Rappaport, N., Schork, N. J., Girke, T., & Monti, S. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.08.14.744937
Sallam, A., & Gillis, J. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.08.22.746357
A transcription factor regulatory atlas for activity inference and perturbation prediction
Sugimoto, H., Tsuyuzaki, K., Zou, Z., Oki, S., Ohta, T., & Kawakami, E. (2026) · Nucleic Acids Research · doi:10.1093/nar/gkag897
AmbiModBench: Benchmarking Gene Perturbation Prediction Beyond Shared ResponsesPreprint
Huang, S., Yang, Z., Yu, K., Chen, J., & Li, S. Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.32527
Self-citationA split-control criterion for few-shot adaptation in single-cell perturbation predictionPreprint
Liang, W., Yang, Y., Zhang, Q.-Y., Feng, J.-C., Wang, Q., Zou, D., Tang, J., Liu, Y., Wang, S., Duan, J., et al. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.10.05.756906 · Cited by 1
Self-citationPlate design confounds donor variation and model evaluation in single-cell perturbation screensPreprint
Ghajari, G., Thirunarayan, K., & Amsaad, F. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.11.04.764088
Single-cell foundation models in perturbation prediction: progress, pitfalls and moving forward
Curean, V., Brown, S. D., Mahale, M., Berindan‐Neagoe, I., Seal, S., Szalay, K. Z., Veres, D. V., & Bender, A. (2026) · Digital Discovery · doi:10.1039/d6dd00209a
Metadata
- Authors
- Li, L., You, Y., Fu, Y., Liao, W., Fan, X., Lu, S., Cao, Y., Li, B., Ren, W., Kong, J., Zheng, S., Chen, J., Liu, X., & Tian, L.
- Affiliations
- Guangzhou Laboratory (CN)
- University of Hong Kong (HK)
- Westlake University (CN)
- Renaissance Era Innovation Technology Co., Ltd., Beijing, China.
- Shanghai Jiao Tong University (CN)
- First Affiliated Hospital of Guangzhou Medical University (CN)
- Corresponding author
- Li, Lanxiang
- You, Yue
- Fu, Yunlin
- Liu, Xiaodong
- Tian, Luyi
- ORCID
- Venue
- Science Advances
- Published
- 2026-09-11
- Volume
- 12
- Issue
- 37
- Pages
- eaed3414-eaed3414
- ISSN
- 2375-2548 (ISSN-L)
- PMC
- PMC13557097
- OpenAlex
- W7212020435
- Cited by
- 1
- References
- 60
- Citations by year
- 2026: 1
- FWCI
- 2.04
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by
- Locations
- Science Advances (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- Funders
- Major Project of Guangzhou National Laboratory (GZNL2024A01015)
- Topics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Single-cell and spatial transcriptomics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene Regulatory Network Analysis
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Chromatin Dynamics
- Keywords
- benchmarking, gene expression regulation, drug response, immune cells, foundation models, virtual cell model, evaluation metrics
- MeSH terms
- Single-Cell Gene Expression Analysis, Animals, Gene Expression Regulation, Humans, Computational Biology, Gene Expression Profiling, Single-Cell Analysis
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.