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.

Mapping single-cell data to reference atlases by transfer learning

Open access

Abstract

Large single-cell atlases are now routinely generated to serve as references for analysis of smaller-scale studies. Yet learning from reference data is complicated by batch effects between datasets, limited availability of computational resources and sharing restrictions on raw data. Here we introduce a deep learning strategy for mapping query datasets on top of a reference called single-cell architectural surgery (scArches). scArches uses transfer learning and parameter optimization to enable efficient, decentralized, iterative reference building and contextualization of new datasets with existing references without sharing raw data. Using examples from mouse brain, pancreas, immune and whole-organism atlases, we show that scArches preserves biological state information while removing batch effects, despite using four orders of magnitude fewer parameters than de novo integration. scArches generalizes to multimodal reference mapping, allowing imputation of missing modalities. Finally, scArches retains coronavirus disease 2019 (COVID-19) disease variation when mapping to a healthy reference, enabling the discovery of disease-specific cell states. scArches will facilitate collaborative projects by enabling iterative construction, updating, sharing and efficient use of reference atlases.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41587-021-01001-7

Cites

84 of 84 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 scArches enables iterative query-to-reference single-cell integration. a, Pre-training of a latent representation using public reference datasets and corresponding reference labels. b, Decentralized model building: users download parameters for the atlas of interest, fine tune the model and optionally upload their updated model for other users. c–e, Illustration of this workflow for a human pancreas atlas across different scArches base models. Training a reference atlas across three human pancreas datasets (CelSeq, InDrop, Fluidigm C1), uniform manifold approximation and projection (UMAP) embedding for the original (c) and the integrated reference for pre-trained reference models (d,e, first column). Second column in d,e, querying a new SS2 dataset to the integrated reference. Updating the cell atlas with a fifth dataset (CelSeq2). Third column in d,e, black dashed circles represent cells absent in the reference data. UMAP plots are based on the model embedding.
Fig. 1 scArches enables iterative query-to-reference single-cell integration. a, Pre-training of a latent representation using public reference datasets and corresponding reference labels. b, Decentralized model building: users download parameters for the atlas of interest, fine tune the model and optionally upload their updated model for other users. c–e, Illustration of this workflow for a human pancreas atlas across different scArches base models. Training a reference atlas across three human pancreas datasets (CelSeq, InDrop, Fluidigm C1), uniform manifold approximation and projection (UMAP) embedding for the original (c) and the integrated reference for pre-trained reference models (d,e, first column). Second column in d,e, querying a new SS2 dataset to the integrated reference. Updating the cell atlas with a fifth dataset (CelSeq2). Third column in d,e, black dashed circles represent cells absent in the reference data. UMAP plots are based on the model embedding.CC BY · PubMed Central
Fig. 2 TL and architecture surgery allow fast and accurate reference mapping. a–c, Comparing different granularity levels in the proposed TL strategy by mapping data from two brain studies to a reference brain atlas. The reference model was trained on a subset of 250,000 cells from two brain studies and then updated with data from Zeisel et al.35 and the TM brain subset. Fine-tuning strategies vary from training a few query study label weights (a) to input layers of both the encoder and decoder (b) and retraining the full network (c). d, Number of trained weights across these three granularity levels. e, Comparison of integration accuracy for different fine-tuning strategies on mapping data from two query studies on a brain reference atlas across various base models. Individual scores are minimum–maximum scaled between 0 and 1. Overall scores were computed using a 40:60-weighted mean of batch correction and bio-conservation scores, respectively (see Methods for further visualization details). EBM, entropy of batch mixing; PCR, principal-component regression.
Fig. 2 TL and architecture surgery allow fast and accurate reference mapping. a–c, Comparing different granularity levels in the proposed TL strategy by mapping data from two brain studies to a reference brain atlas. The reference model was trained on a subset of 250,000 cells from two brain studies and then updated with data from Zeisel et al.35 and the TM brain subset. Fine-tuning strategies vary from training a few query study label weights (a) to input layers of both the encoder and decoder (b) and retraining the full network (c). d, Number of trained weights across these three granularity levels. e, Comparison of integration accuracy for different fine-tuning strategies on mapping data from two query studies on a brain reference atlas across various base models. Individual scores are minimum–maximum scaled between 0 and 1. Overall scores were computed using a 40:60-weighted mean of batch correction and bio-conservation scores, respectively (see Methods for further visualization details). EBM, entropy of batch mixing; PCR, principal-component regression.CC BY · PubMed Central
Fig. 3 scArches enables efficient reference mapping compared to full integration workflow with existing data-integration methods. a–d, Evaluating the effect of the reference (ref) size for immune data (n = 20,522) on the quality of reference mapping. a–c, UMAP plots show the latent representation of the integrated query and reference data together for scArches scANVI. Cell types highlighted with dashed circles represent cells unique to a specific study denoted by the batch legend. Reference ratio refers to the fraction of cells in the reference compared to all data. The studies used as reference are indicated at the bottom of each panel. HSPCs, hematopoietic stem and progenitor cells. d, Quantitative metrics for the performance of different base models across various reference ratios in immune data. e, Comparison of different scArches base models trained with a reference dataset with ~66% of batches in the whole data against de novo full integration methods across immune (n = 20,522), pancreas (n = 15,681) and brain (n = 332,129) datasets. Conos and mnnCorrect were not able to integrate brain data due to excessive memory usage and time requirements, respectively. MSE, mean squared error.
Fig. 3 scArches enables efficient reference mapping compared to full integration workflow with existing data-integration methods. a–d, Evaluating the effect of the reference (ref) size for immune data (n = 20,522) on the quality of reference mapping. a–c, UMAP plots show the latent representation of the integrated query and reference data together for scArches scANVI. Cell types highlighted with dashed circles represent cells unique to a specific study denoted by the batch legend. Reference ratio refers to the fraction of cells in the reference compared to all data. The studies used as reference are indicated at the bottom of each panel. HSPCs, hematopoietic stem and progenitor cells. d, Quantitative metrics for the performance of different base models across various reference ratios in immune data. e, Comparison of different scArches base models trained with a reference dataset with ~66% of batches in the whole data against de novo full integration methods across immune (n = 20,522), pancreas (n = 15,681) and brain (n = 332,129) datasets. Conos and mnnCorrect were not able to integrate brain data due to excessive memory usage and time requirements, respectively. MSE, mean squared error.CC BY · PubMed Central
Fig. 4 scArches successfully transfers knowledge from reference to query. a,b, Querying TM (n = 90,120) to the larger reference atlas Tabula Senis (n = 264,287) using scArches trVAE (a) across different tissues (b). Tissues were correctly grouped across the two datasets (a,b). c, Location of misclassified and unknown cells after transferring labels from the reference to the query data. The highlighted tissue represents tracheal cells, which we removed from the reference data. d, Reported uncertainty of the transferred labels, which was low in correctly classified cells and high in the incorrect and unknown ones, particularly in the trachea. Box plots indicate the median (center lines) and interquartile range (hinges), and whiskers represents minimum and maximum values. Numbers of cells (n) are denoted above each box plot. e, Numbers of correct, incorrect and unknown cells across different tissues. The red dashed line represents tracheal cells only present in TM. f, Construction of a reference CITE-seq atlas using two PBMC datasets (n = 10,849 cells). g, Integration of scRNA-seq data (n = 10,315) into the CITE-seq reference. h, Imputation of missing proteins for the query dataset using the reference. BAT, brown adipose tissue; GAT, gonadal adipose tissue; MAT, mesenteric adipose tissue; SCAT, subcutaneous adipose tissue.
Fig. 4 scArches successfully transfers knowledge from reference to query. a,b, Querying TM (n = 90,120) to the larger reference atlas Tabula Senis (n = 264,287) using scArches trVAE (a) across different tissues (b). Tissues were correctly grouped across the two datasets (a,b). c, Location of misclassified and unknown cells after transferring labels from the reference to the query data. The highlighted tissue represents tracheal cells, which we removed from the reference data. d, Reported uncertainty of the transferred labels, which was low in correctly classified cells and high in the incorrect and unknown ones, particularly in the trachea. Box plots indicate the median (center lines) and interquartile range (hinges), and whiskers represents minimum and maximum values. Numbers of cells (n) are denoted above each box plot. e, Numbers of correct, incorrect and unknown cells across different tissues. The red dashed line represents tracheal cells only present in TM. f, Construction of a reference CITE-seq atlas using two PBMC datasets (n = 10,849 cells). g, Integration of scRNA-seq data (n = 10,315) into the CITE-seq reference. h, Imputation of missing proteins for the query dataset using the reference. BAT, brown adipose tissue; GAT, gonadal adipose tissue; MAT, mesenteric adipose tissue; SCAT, subcutaneous adipose tissue.CC BY · PubMed Central
Fig. 5 scArches resolves severity in COVID-19 query data mapped to a healthy reference and reveals emergent cell states. a–c, Integration of query data from immune and epithelial cells from patients with COVID-19 on top of a healthy immune atlas across multiple tissues (a), cell types (b) and cell states (c). BALF, bronchoalveolar lavage fluid; DC, dendritic cell; Treg cells, regulatory T cells. d, Comparison of various macrophage subpopulations across both healthy and COVID-19 states. Top, TRAMs are characterized by expression of FABP4, while monocyte-derived inflammatory macrophages (MoMs) are characterized by expression of CCL2. Upregulation of C1QA illustrates maturation of MoMs as they differentiate from monocytes to macrophages. Middle, CXCL5, IFI27 and CXCL10 illustrate context-dependent activation of TRAMs. Bottom, scArches correctly maps TRAMs from query to TRAMs from reference, while preserving MoMs, unseen in the reference, as a distinct cell type. e, Separation of activated query CD8+ T cells from patients with COVID-19 from the rest of CD8+ T cells in the reference. AT, alveolar type; mDC, myeloid dendritic cells; pDC, plasmacytoid dendritic cells.
Fig. 5 scArches resolves severity in COVID-19 query data mapped to a healthy reference and reveals emergent cell states. a–c, Integration of query data from immune and epithelial cells from patients with COVID-19 on top of a healthy immune atlas across multiple tissues (a), cell types (b) and cell states (c). BALF, bronchoalveolar lavage fluid; DC, dendritic cell; Treg cells, regulatory T cells. d, Comparison of various macrophage subpopulations across both healthy and COVID-19 states. Top, TRAMs are characterized by expression of FABP4, while monocyte-derived inflammatory macrophages (MoMs) are characterized by expression of CCL2. Upregulation of C1QA illustrates maturation of MoMs as they differentiate from monocytes to macrophages. Middle, CXCL5, IFI27 and CXCL10 illustrate context-dependent activation of TRAMs. Bottom, scArches correctly maps TRAMs from query to TRAMs from reference, while preserving MoMs, unseen in the reference, as a distinct cell type. e, Separation of activated query CD8+ T cells from patients with COVID-19 from the rest of CD8+ T cells in the reference. AT, alveolar type; mDC, myeloid dendritic cells; pDC, plasmacytoid dendritic cells.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

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.

The self-check needs the paper’s full text.

Citations

Cites

Show all 76 references

Research works printed but not linked: 2

  1. Stuart, T. et al. Comprehensive integration of single cell data. Cell 177, 1888–1902 (2019). · DOI · Paper page
  2. Sohn, K., Lee, H. & Yan, X. Learning structured output representation using deep conditional generative models. in Advances in Neural Information Processing Systems (eds. Cortes, C. et al.) 28, 3483–3491 (Curran Associates, 2015).

Cited by

See all 613

Metadata

Authors
Lotfollahi, M., Naghipourfar, M., Luecken, M. D., Khajavi, M., Büttner, M., Wagenstetter, M., Avsec, Ž., Gayoso, A., Yosef, N., Interlandi, M., Rybakov, S., Misharin, A. V., & Theis, F. J.
Affiliations
  • Helmholtz Munich (DE)
  • Technical University of Munich (DE)
  • University of California, Berkeley (US)
  • Ragon Institute of MGH, MIT and Harvard (US)
  • University of Münster (DE)
  • Northwestern University (US)
Corresponding author
Theis, Fabian J.
Venue
Nature Biotechnology
Published
2021-08-30
Volume
40
Issue
1
Pages
121-130
ISSN
  • 0733-222X (ISSN-L)
  • 1087-0156
  • 1546-1696
OpenAlex
W3196650609
Cited by
602
References
76
Citations by year
2026: 130 · 2025: 144 · 2024: 146 · 2023: 116 · 2022: 57 · 2021: 7
FWCI
26.29
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Biotechnology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PuSH - Publication Server of Helmholtz Zentrum München (submittedVersion, cc-by)
  • eScholarship (California Digital Library) (submittedVersion)
  • mediaTUM (Technical University of Munich) (submittedVersion, other-oa)
  • Site cant be reached (submittedVersion, cc-by)
Funders
Deutsche Forschungsgemeinschaft (ZT-I-0007)
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 › Bioinformatics and Genomic Networks
Keywords
transfer learning, single-cell data integration, batch effect correction, mouse brain atlas
MeSH terms
Deep Learning, COVID-19, SARS-CoV-2, Animals, Humans, Organ Specificity, Reference Standards, Mice, Single-Cell Analysis, Datasets as Topic

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.

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: Mapping single-cell data to reference atlases by transfer learning. https://app.cochat.ai/papers/10.1038/s41587-021-01001-7
The paper
Lotfollahi, M., Naghipourfar, M., Luecken, M. D., Khajavi, M., Büttner, M., Wagenstetter, M., Avsec, Ž., Gayoso, A., Yosef, N., Interlandi, M., Rybakov, S., Misharin, A. V., & Theis, F. J. (2022). Mapping single-cell data to reference atlases by transfer learning. Nature Biotechnology. https://doi.org/10.1038/s41587-021-01001-7

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.