Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 84 of the 84 references it cites were found in a registry or an index; 19 of the 39 sources its claims cite were checked for support.

Causal identification of single-cell experimental perturbation effects with CINEMA-OT

Open access
Dong, Mingze, Wang, Bao, Wei, Jessica, de O. Fonseca, Antonio H., Perry, Curtis J., Frey, Alexander, Ouerghi, Feriel, Foxman, Ellen F., Ishizuka, Jeffrey J., Dhodapkar, Rahul M., van Dijk, David · 2023 · Nature Methods

Published version of a preprint

Abstract

Recent advancements in single-cell technologies allow characterization of experimental perturbations at single-cell resolution. While methods have been developed to analyze such experiments, the application of a strict causal framework has not yet been explored for the inference of treatment effects at the single-cell level. Here we present a causal-inference-based approach to single-cell perturbation analysis, termed CINEMA-OT (causal independent effect module attribution + optimal transport). CINEMA-OT separates confounding sources of variation from perturbation effects to obtain an optimal transport matching that reflects counterfactual cell pairs. These cell pairs represent causal perturbation responses permitting a number of novel analyses, such as individual treatment-effect analysis, response clustering, attribution analysis, and synergy analysis. We benchmark CINEMA-OT on an array of treatment-effect estimation tasks for several simulated and real datasets and show that it outperforms other single-cell perturbation analysis methods. Finally, we perform CINEMA-OT analysis of two newly generated datasets: (1) rhinovirus and cigarette-smoke-exposed airway organoids, and (2) combinatorial cytokine stimulation of immune cells. In these experiments, CINEMA-OT reveals potential mechanisms by which cigarette-smoke exposure dulls the airway antiviral response, as well as the logic that governs chemokine secretion and peripheral immune cell recruitment.

Overview

CoChat Summary

TL;DR

CINEMA-OT is a causal-inference framework for single-cell perturbation analysis that separates confounding variation from treatment-associated variation using ICA and then matches counterfactual cell pairs via entropy-regularized optimal transport. It benchmarks favorably against existing methods and reveals heterogeneous, synergistic responses in airway organoids exposed to rhinovirus and cigarette smoke and in cytokine-stimulated immune cells.

Key findings

  • CINEMA-OT identifies confounder signals with ICA and a Chatterjee’s coefficient-based test, then uses weighted optimal transport to generate causal counterfactual cell pairs and individual treatment effects (ITEs).
  • In simulated and real benchmarks, CINEMA-OT outperformed Mixscape, Harmony-Mixscape, scGen, CPA, ContrastiveVI, CellOT, and Full OT in treatment-effect attribution and batch mixing, while preserving differential-response clusters; it ran in approximately 1 min for 50,000 cells and used less than 12 GB memory.
  • In human airway organoids (26,420 cells across mock, RV, CSE, and RVCSE), CINEMA-OT found that cigarette-smoke exposure broadly reduced interferon-stimulated gene responses to rhinovirus, while pre-ciliated cells showed synergistic induction of secretory genes such as SCGB3A1, LCN2, BPIFB1, SLPI, and WFDC2.
  • In combinatorial cytokine stimulation of PBMCs, CINEMA-OT identified monocyte-enriched synergistic chemokine programs, including inhibition of baseline neutrophil chemotactic signaling and induction of monocyte chemotactic signaling with IFN-β + IL-6, and lymphocyte-predominant chemokines with added IFN-γ.

Methods

  • CINEMA-OT is an unsupervised method: PCA embedding, FastICA decomposition, Chatterjee’s cross-rank coefficient to separate confounder versus treatment-associated independent components, and entropy-regularized optimal transport solved by Sinkhorn–Knopp to match cells across conditions.
  • Benchmarks used simulated scRNA-seq data (1,000 genes × 5,000 cells with 2–5 cell states), the Alzheimer’s entorhinal cortex dataset, and the sci-Plex drug perturbation dataset; comparisons included Mixscape, Harmony-Mixscape, scGen, CPA, ContrastiveVI, CellOT, and Full OT.
  • New datasets included primary human bronchial organoids infected with mock or rhinovirus with/without 2% cigarette-smoke extract (26,420 cells, four samples) and PBMCs from three healthy donors stimulated acutely (2 d) or chronically (7 d) with type I, II, and III IFNs and combinations with TNF or IL-6.

Limitations

  • Stated: CINEMA-OT faces bias–variance trade-offs in optimal transport smoothness; too much smoothing can create false positives, while too little harms stability and increases variance.
  • Stated: CINEMA-OT cannot extrapolate to unmeasured perturbation–cell pairs, is not designed for cases where changes in confounder distributions are the primary effect of interest, and its current implementation does not perform count modeling, so gene-level differential expression may still be affected by substantial batch effects.
  • Stated: CINEMA-OT-W addresses differential abundance but its theoretical foundation no longer holds under differential abundance, and selecting its clustering resolution may require prior biological knowledge; the text also notes that CINEMA-OT can tolerate only moderate differential abundance without CINEMA-OT-W.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41592-023-02040-5

Cites

84 of 84 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Overview of the CINEMA-OT framework. a, scRNA-seq count data is first decomposed into confounder variation and treatment-associated variation using ICA. Cells are then matched across treatment conditions by entropy-regularized optimal transport in the confounder space to generate a causal matching plan. The smooth matching map can then be used to estimate individual treatment effects. b, Illustration of the properties of CINEMA-OT compared with other potential matching schemes, including global matching (minimizing the average difference) and local matching (finding nearest neighbors for each cell). c, Illustration of the differential abundance issue in the unweighted CINEMA-OT method, and the resampling procedure used in CINEMA-OT-W.
Fig. 1 Overview of the CINEMA-OT framework. a, scRNA-seq count data is first decomposed into confounder variation and treatment-associated variation using ICA. Cells are then matched across treatment conditions by entropy-regularized optimal transport in the confounder space to generate a causal matching plan. The smooth matching map can then be used to estimate individual treatment effects. b, Illustration of the properties of CINEMA-OT compared with other potential matching schemes, including global matching (minimizing the average difference) and local matching (finding nearest neighbors for each cell). c, Illustration of the differential abundance issue in the unweighted CINEMA-OT method, and the resampling procedure used in CINEMA-OT-W.CC BY · PubMed Central
Fig. 2 Key functionalities of CINEMA-OT. a, CINEMA-OT takes scRNA-seq data labeled by treatment condition as input. CINEMA-OT learns a confounder embedding that is mixed across batches and matches counterfactual cell pairs across conditions to compute causal perturbation effects. b, The single-cell-level treatment-effect matrices can be further clustered, and gene set enrichment analysis can be conducted on the output. GO, Gene Ontology. c, Single-cell-level synergy in combinatorial perturbations can be obtained as the dissimilarity of extrapolated phenotypes and true combinatorially perturbed phenotypes. d, CINEMA-OT can attribute divergent treatment effects to either explicit confounders or latent confounders by analysis of cluster-wise response matching matrices.
Fig. 2 Key functionalities of CINEMA-OT. a, CINEMA-OT takes scRNA-seq data labeled by treatment condition as input. CINEMA-OT learns a confounder embedding that is mixed across batches and matches counterfactual cell pairs across conditions to compute causal perturbation effects. b, The single-cell-level treatment-effect matrices can be further clustered, and gene set enrichment analysis can be conducted on the output. GO, Gene Ontology. c, Single-cell-level synergy in combinatorial perturbations can be obtained as the dissimilarity of extrapolated phenotypes and true combinatorially perturbed phenotypes. d, CINEMA-OT can attribute divergent treatment effects to either explicit confounders or latent confounders by analysis of cluster-wise response matching matrices.CC BY · PubMed Central
Fig. 3 Benchmarking of CINEMA-OT against other methods for single-cell perturbation analysis. a, Illustrations of the data stimulation and metrics included in the benchmarking. b, Illustration of our conducted simulation study. c, Box plots of different validation metrics on synthetic data for CINEMA-OT and other methods (n = 15 for confounder embedding metrics, n = 12 for ITE metrics). ITE metrics were computed only for datasets with differential responses. The top and bottom hinges represent the top and bottom quartiles, and whiskers extend from the hinge to the largest or smallest value no further than 1.5 × the interquartile range from the hinge. The median is used as the center. ASW, Average silhouette width. ARI, adjusted Rand index. d, Comparison of the performance of different methods across synthetic datasets with various differential-abundance ratio settings: 1, 0.75, 0.5, 0.25, 0 (missing cell types), with n = 15 for confounder embedding metrics and n = 12 for ITE metrics in each setting. Data are presented as mean values ± s.d. e, Illustration of the validation of CINEMA-OT on the Sciplex dataset. f, Quantification of different validation metrics on the Sciplex dataset for CINEMA-OT and other methods.
Fig. 3 Benchmarking of CINEMA-OT against other methods for single-cell perturbation analysis. a, Illustrations of the data stimulation and metrics included in the benchmarking. b, Illustration of our conducted simulation study. c, Box plots of different validation metrics on synthetic data for CINEMA-OT and other methods (n = 15 for confounder embedding metrics, n = 12 for ITE metrics). ITE metrics were computed only for datasets with differential responses. The top and bottom hinges represent the top and bottom quartiles, and whiskers extend from the hinge to the largest or smallest value no further than 1.5 × the interquartile range from the hinge. The median is used as the center. ASW, Average silhouette width. ARI, adjusted Rand index. d, Comparison of the performance of different methods across synthetic datasets with various differential-abundance ratio settings: 1, 0.75, 0.5, 0.25, 0 (missing cell types), with n = 15 for confounder embedding metrics and n = 12 for ITE metrics in each setting. Data are presented as mean values ± s.d. e, Illustration of the validation of CINEMA-OT on the Sciplex dataset. f, Quantification of different validation metrics on the Sciplex dataset for CINEMA-OT and other methods.CC BY · PubMed Central
Fig. 4 CINEMA-OT identifies a heterogeneous defensive response in human airway epithelial cells exposed to rhinovirus and cigarette-smoke extract. a, Overview of experimental design. Differentiated airway epithelial organoids are challenged with mock (control) or RV infection, with or without CSE exposure. b, UMAP projection of expression data labeled by perturbation and cell type. c, UMAP projection of the individual treatment-effect matrix obtained by CINEMA-OT from the RV response without CSE exposure, colored by response cluster. d, Gene set enrichment analysis of mock to RV response clusters identified by CINEMA-OT, colored by adjusted P value. e, UMAP projection of the individual treatment effect matrix obtained by CINEMA-OT from the CSE response without RV exposure, colored by response cluster. f, Gene set enrichment analysis of mock to CSE response clusters identified by CINEMA-OT, colored by adjusted P value. g, Cell-wise synergy score visualization. h, UMAP visualization of CINEMA-OT synergy embedding, colored by cell type. f, Two diverging patterns among CINEMA-OT identified synergistic genes, visualized by stacked violin plots and dot plots.
Fig. 4 CINEMA-OT identifies a heterogeneous defensive response in human airway epithelial cells exposed to rhinovirus and cigarette-smoke extract. a, Overview of experimental design. Differentiated airway epithelial organoids are challenged with mock (control) or RV infection, with or without CSE exposure. b, UMAP projection of expression data labeled by perturbation and cell type. c, UMAP projection of the individual treatment-effect matrix obtained by CINEMA-OT from the RV response without CSE exposure, colored by response cluster. d, Gene set enrichment analysis of mock to RV response clusters identified by CINEMA-OT, colored by adjusted P value. e, UMAP projection of the individual treatment effect matrix obtained by CINEMA-OT from the CSE response without RV exposure, colored by response cluster. f, Gene set enrichment analysis of mock to CSE response clusters identified by CINEMA-OT, colored by adjusted P value. g, Cell-wise synergy score visualization. h, UMAP visualization of CINEMA-OT synergy embedding, colored by cell type. f, Two diverging patterns among CINEMA-OT identified synergistic genes, visualized by stacked violin plots and dot plots.CC BY · PubMed Central
Fig. 5 CINEMA-OT reveals combinatorial mechanisms of acute and chronic cytokine stimulation. a, Illustration of experimental design. b, UMAP projection of expression data colored by samples, perturbations, and cell types. In sample labels, H refers to the donor number, and D refers to the number of days of stimulation. NK, natural killer. c, UMAP projection of the CD4+ T cell counterfactual space from CINEMA-OT. Projections are colored by experimental batch and perturbation type. d, UMAP projection of the monocyte counterfactual space from CINEMA-OT. Projections are colored by experimental batch and perturbation type. e, UMAP projections of the original data, confounder embedding, and individual treatment effects identified by CINEMA-OT after acute stimulation with IFN-β in H3D2, colored by response cluster and cell type. f, Volcano plot highlighting genes with strong confounder-specific treatment effects. g, Normalized expression of representative confounder-specific treatment-associated genes in original UMAP space. h, Distribution of gene synergy score, obtained by combining results of IFN-β and TNF, IFN-β and IFN-γ, and IFN-β and IL-6 treatments in H3D2. i, Cell-wise synergy score visualization in the acute condition, taking a single experimental batch (H3D2) as an example. j, Stacked gene expression violin plot of synergistic chemokines identified by CINEMA-OT. k, Patterns of chemokine secretion programmed by single or multi-signal cytokine stimulation. X: TNF, IFN-γ and IL-6.
Fig. 5 CINEMA-OT reveals combinatorial mechanisms of acute and chronic cytokine stimulation. a, Illustration of experimental design. b, UMAP projection of expression data colored by samples, perturbations, and cell types. In sample labels, H refers to the donor number, and D refers to the number of days of stimulation. NK, natural killer. c, UMAP projection of the CD4+ T cell counterfactual space from CINEMA-OT. Projections are colored by experimental batch and perturbation type. d, UMAP projection of the monocyte counterfactual space from CINEMA-OT. Projections are colored by experimental batch and perturbation type. e, UMAP projections of the original data, confounder embedding, and individual treatment effects identified by CINEMA-OT after acute stimulation with IFN-β in H3D2, colored by response cluster and cell type. f, Volcano plot highlighting genes with strong confounder-specific treatment effects. g, Normalized expression of representative confounder-specific treatment-associated genes in original UMAP space. h, Distribution of gene synergy score, obtained by combining results of IFN-β and TNF, IFN-β and IFN-γ, and IFN-β and IL-6 treatments in H3D2. i, Cell-wise synergy score visualization in the acute condition, taking a single experimental batch (H3D2) as an example. j, Stacked gene expression violin plot of synergistic chemokines identified by CINEMA-OT. k, Patterns of chemokine secretion programmed by single or multi-signal cytokine stimulation. X: TNF, IFN-γ and IL-6.CC BY · PubMed Central
Extended Data Fig. 1 CINEMA-OT validation on a simulated dataset example with differential abundance. A-G. UMAP visualizations of confounder space (left) and treatment effect space (right), both colored by cell type ("states") and response patterns ("Responses") in a simulated dataset with differential abundance ratio = 0.25.
Extended Data Fig. 1 CINEMA-OT validation on a simulated dataset example with differential abundance. A-G. UMAP visualizations of confounder space (left) and treatment effect space (right), both colored by cell type ("states") and response patterns ("Responses") in a simulated dataset with differential abundance ratio = 0.25.CC BY · PubMed Central
Extended Data Fig. 2 Comparison for different single-cell level treatment effect analysis methods’ performance as the sparsity level of the dataset increases. Here the result of scGen and ContrastiveVI in the setting of subsampling level = 0.95 is not shown, as they fail to converge or return NaN in some datasets. Data are presented as mean values +/- std (n=15 for confounder embedding metrics, n=12 for ITE metrics). The smaller number of n for ITE metrics is because they are only meaningful for datasets with response pattern numbers larger than 1.
Extended Data Fig. 2 Comparison for different single-cell level treatment effect analysis methods’ performance as the sparsity level of the dataset increases. Here the result of scGen and ContrastiveVI in the setting of subsampling level = 0.95 is not shown, as they fail to converge or return NaN in some datasets. Data are presented as mean values +/- std (n=15 for confounder embedding metrics, n=12 for ITE metrics). The smaller number of n for ITE metrics is because they are only meaningful for datasets with response pattern numbers larger than 1.CC BY · PubMed Central
Extended Data Fig. 3 Running time and peak memory usage comparison for different single-cell level treatment effect analysis methods. *: These methods are tested on high performance clusters, therefore the performance may not be directly comparable. Data are presented as mean values +/- std. The sample size equals 5 in each setting with the exception of scGen, CPA, and ContrastiveVI(GPU) with ncells = 50000, in which settings the sample size is set to 2 to save time.
Extended Data Fig. 3 Running time and peak memory usage comparison for different single-cell level treatment effect analysis methods. *: These methods are tested on high performance clusters, therefore the performance may not be directly comparable. Data are presented as mean values +/- std. The sample size equals 5 in each setting with the exception of scGen, CPA, and ContrastiveVI(GPU) with ncells = 50000, in which settings the sample size is set to 2 to save time.CC BY · PubMed Central
Extended Data Fig. 4 CINEMA-OT validation on Alzheimer’s data. A. UMAP visualizations of different covariates (Cell type, donor batch, and AD/control) and SPP1 expression. B-I. Different methods’ confounder space visualization and treatment effect visualization, colored by cell type and condition (for confounder space), cell type and SPP1 expression (for response space) respectively.
Extended Data Fig. 4 CINEMA-OT validation on Alzheimer’s data. A. UMAP visualizations of different covariates (Cell type, donor batch, and AD/control) and SPP1 expression. B-I. Different methods’ confounder space visualization and treatment effect visualization, colored by cell type and condition (for confounder space), cell type and SPP1 expression (for response space) respectively.CC BY · PubMed Central
Extended Data Fig. 5 Additional comparisons of ContrastiveVI and CINEMA-OT in the Alzheimer scRNA-seq dataset. A. UMAP visualizations of different covariates (Cell type, donor sex, donor batch, and AD/control) and expressions of different cell-type specific AD marker genes. B. ContrastiveVI response space visualization for covariates and leiden subclusters (first line), state-specific gene visualization in the response space (second line), and confounder space visualization for covariates (third line). C. CINEMA-OT response space visualization for covariates and Leiden subclusters (first line), state-specific gene visualization in the response space (second line), and confounder space visualization for covariates (third line). D. Response-cluster specific genes identified by CINEMA-OT.
Extended Data Fig. 5 Additional comparisons of ContrastiveVI and CINEMA-OT in the Alzheimer scRNA-seq dataset. A. UMAP visualizations of different covariates (Cell type, donor sex, donor batch, and AD/control) and expressions of different cell-type specific AD marker genes. B. ContrastiveVI response space visualization for covariates and leiden subclusters (first line), state-specific gene visualization in the response space (second line), and confounder space visualization for covariates (third line). C. CINEMA-OT response space visualization for covariates and Leiden subclusters (first line), state-specific gene visualization in the response space (second line), and confounder space visualization for covariates (third line). D. Response-cluster specific genes identified by CINEMA-OT.CC BY · PubMed Central
Extended Data Fig. 6 CINEMA-OT validation on Sciplex data. A. Schematic overview of the benchmarking design, same as the overview shown in Fig. 3e. B. The bar plot showing metric comparison results across benchmarked methods, same as Fig. 3f. C-I. UMAP visualization of different methods’ confounder space, colored by the perturbation level of Pracinostat, Acetate, Citrate, Pyruvate, and the defined confounder states respectively.
Extended Data Fig. 6 CINEMA-OT validation on Sciplex data. A. Schematic overview of the benchmarking design, same as the overview shown in Fig. 3e. B. The bar plot showing metric comparison results across benchmarked methods, same as Fig. 3f. C-I. UMAP visualization of different methods’ confounder space, colored by the perturbation level of Pracinostat, Acetate, Citrate, Pyruvate, and the defined confounder states respectively.CC BY · PubMed Central
Extended Data Fig. 7 Visualizations of identified genes with strong confounder-dependent treatment effect, following Fig. 5f. UMAP visualizations of identified genes with strong confounder-dependent treatment effect, colored by perturbation condition, cell type annotation, and log normalized gene expression.
Extended Data Fig. 7 Visualizations of identified genes with strong confounder-dependent treatment effect, following Fig. 5f. UMAP visualizations of identified genes with strong confounder-dependent treatment effect, colored by perturbation condition, cell type annotation, and log normalized gene expression.CC BY · PubMed Central
Extended Data Fig. 8 CINEMA-OT validation on the interferon data. A-H. Different methods’ confounder space visualization (colored by treatment condition and cell types) and treatment effect visualization (colored by cell types and Leiden clusters of the response space).
Extended Data Fig. 8 CINEMA-OT validation on the interferon data. A-H. Different methods’ confounder space visualization (colored by treatment condition and cell types) and treatment effect visualization (colored by cell types and Leiden clusters of the response space).CC BY · PubMed Central
Extended Data Fig. 9 Comparison of CD4 T cells gene expression change between chronic stimulation condition and acute stimulation condition across different conditions. A. UMAP visualizations of CD4 T cell response space pooled over different perturbations, colored by donor and perturbation condition. B. Dot plots of selected genes identified by CINEMA-OT. C. Single-cell expression differences of selected genes visualized in the UMAP space.
Extended Data Fig. 9 Comparison of CD4 T cells gene expression change between chronic stimulation condition and acute stimulation condition across different conditions. A. UMAP visualizations of CD4 T cell response space pooled over different perturbations, colored by donor and perturbation condition. B. Dot plots of selected genes identified by CINEMA-OT. C. Single-cell expression differences of selected genes visualized in the UMAP space.CC BY · PubMed Central

Verification

Claims the paper makes

10 Valid · 2 Not valid · 11 Not checked · 88 Own results and interpretations

45 more claims · 2 with their cited sources checked

Create a free account to see all 111 claims and to verify the 20 references still open.

See all claims — free account
  • Probably not a substantial claim. 35
  • It may state two separate claims as one. 35

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

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.

3 of 3

Citations

Cites

Show all 81 references

Research works printed but not linked: 2

  1. Cuturi, M. Sinkhorn distances: lightspeed computation of optimal transport. In Advances in Neural Information Processing Systems Vol. 26 (eds Burges, C. J. et al.) (2013).
  2. Dong, M. & Kluger, Y. GEASS: neural causal feature selection for high-dimensional biological data. In 11th International Conference on Learning Representations (ICLR, 2023).

Cited by

See all 135

Metadata

Authors
Dong, M., Wang, B., Wei, J., de O. Fonseca, A. H., Perry, C. J., Frey, A., Ouerghi, F., Foxman, E. F., Ishizuka, J. J., Dhodapkar, R. M., & van Dijk, D.
Affiliations
  • Yale University (US)
  • University of Southern California (US)
Corresponding author
  • Foxman, Ellen F.
  • Ishizuka, Jeffrey J.
  • Dhodapkar, Rahul M.
Venue
Nature Methods
Published
2023-11
Volume
20
Issue
11
Pages
1769-1779
ISSN
  • 1548-7091 (ISSN-L)
  • 1548-7105
OpenAlex
W4388217091
Cited by
94
References
81
Citations by year
2026: 37 · 2025: 33 · 2024: 17 · 2023: 7
FWCI
8.35
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Methods (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
  • U.S. Department of Health & Human Services | NIH | National Institute of General Medical Sciences (5R35GM143072)
  • U.S. Department of Health & Human Services | NIH | National Institute of Allergy and Infectious Diseases (3R01AI157488)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Single-cell and spatial transcriptomics
  • Physical Sciences › Mathematics › Statistics and Probability › Advanced Causal Inference Techniques
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene Regulatory Network Analysis
Keywords
causal inference, optimal transport, confounding, treatment effect estimation, synergy analysis, attribution analysis, airway organoids, rhinovirus, cigarette smoke, chemokine secretion, immune cell recruitment, cytokine induction
MeSH terms
Motion Pictures, Cytokines

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 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash and 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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Causal identification of single-cell experimental perturbation effects with CINEMA-OT. https://app.cochat.ai/papers/10.1038/s41592-023-02040-5
The paper
Dong, M., Wang, B., Wei, J., de O. Fonseca, A. H., Perry, C. J., Frey, A., Ouerghi, F., Foxman, E. F., Ishizuka, J. J., Dhodapkar, R. M., & van Dijk, D. (2023). Causal identification of single-cell experimental perturbation effects with CINEMA-OT. Nature Methods. https://doi.org/10.1038/s41592-023-02040-5

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.