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

The DOI matches its Crossref record; 66 of the 66 references it cites were found in a registry or an index.

scDeepSort: a pre-trained cell-type annotation method for single-cell transcriptomics using deep learning with a weighted graph neural network

Open access

Published version of a preprint

Abstract

Advances in single-cell RNA sequencing (scRNA-seq) have furthered the simultaneous classification of thousands of cells in a single assay based on transcriptome profiling. In most analysis protocols, single-cell type annotation relies on marker genes or RNA-seq profiles, resulting in poor extrapolation. Still, the accurate cell-type annotation for single-cell transcriptomic data remains a great challenge. Here, we introduce scDeepSort (https://github.com/ZJUFanLab/scDeepSort), a pre-trained cell-type annotation tool for single-cell transcriptomics that uses a deep learning model with a weighted graph neural network (GNN). Using human and mouse scRNA-seq data resources, we demonstrate the high performance and robustness of scDeepSort in labeling 764 741 cells involving 56 human and 32 mouse tissues. Significantly, scDeepSort outperformed other known methods in annotating 76 external test datasets, reaching an 83.79% accuracy across 265 489 cells in humans and mice. Moreover, we demonstrate the universality of scDeepSort using more challenging datasets and using references from different scRNA-seq technology. Above all, scDeepSort is the first attempt to annotate cell types of scRNA-seq data with a pre-trained GNN model, which can realize the accurate cell-type annotation without additional references, i.e. markers or RNA-seq profiles.

Overview

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

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

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

Integrity signals

Registry

Crossref · 10.1093/nar/gkab775

Cites

66 of 66 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1. General conceptual framework and validation of scDeepSort. (A) Human and mouse single-cell transcriptomics atlases were curated from HCL and MCA as the underlying data for training scDeepSort. The human and mouse atlases include 562,977 cells from 56 tissues and 201,764 cells from 32 tissues, respectively. (B) For each cell, a graph network was constructed of this cell, its genes and neighboring cells for supervised learning scDeepSort with known cell labels from the transcriptomic atlases for each tissue. (C) Internal human and mouse atlases datasets and external test datasets of single-cell transcriptomics data involving multiple tissues were employed to test the performance of scDeepSort. Markers- and profiles-dependent annotation methods (CellAssign, Garnett, SingleR, scMap, ACTINN, CHETAH, scID, scPred, SCINA, singleCellNet and SVM) were compared with scDeepSort on human and mouse external test datasets.
Figure 1. General conceptual framework and validation of scDeepSort. (A) Human and mouse single-cell transcriptomics atlases were curated from HCL and MCA as the underlying data for training scDeepSort. The human and mouse atlases include 562,977 cells from 56 tissues and 201,764 cells from 32 tissues, respectively. (B) For each cell, a graph network was constructed of this cell, its genes and neighboring cells for supervised learning scDeepSort with known cell labels from the transcriptomic atlases for each tissue. (C) Internal human and mouse atlases datasets and external test datasets of single-cell transcriptomics data involving multiple tissues were employed to test the performance of scDeepSort. Markers- and profiles-dependent annotation methods (CellAssign, Garnett, SingleR, scMap, ACTINN, CHETAH, scID, scPred, SCINA, singleCellNet and SVM) were compared with scDeepSort on human and mouse external test datasets.CC BY-NC · PubMed Central
Figure 2. The weighted GNN algorithm of scDeepSort. The algorithm consists of embedding, weighted graph aggregator and linear classifier layers. The embedding layer stores the graph node representations and is frozen during training, wherein dimensionality reduction methods (PCA and weighted sums) were used to generate the initial fixed-size node representations, and the gene expression for each cell was regarded as the weighted edge between cells and genes forming a weighted adjacent matrix. In the weighted aggregator layer, a self-loop confidence for each cell node and a learnable sharing confidence for each gene node were incorporated into the weighted cell–gene graph. For each node (i.e. cell-centered and gene-centered) subgraph, weighted edges were normalized and the node itself and its neighborhood were then gathered to generate a new cell node representation during aggregation. The linear classifier layer categorizes the final cell state representation as a predefined cell type.
Figure 2. The weighted GNN algorithm of scDeepSort. The algorithm consists of embedding, weighted graph aggregator and linear classifier layers. The embedding layer stores the graph node representations and is frozen during training, wherein dimensionality reduction methods (PCA and weighted sums) were used to generate the initial fixed-size node representations, and the gene expression for each cell was regarded as the weighted edge between cells and genes forming a weighted adjacent matrix. In the weighted aggregator layer, a self-loop confidence for each cell node and a learnable sharing confidence for each gene node were incorporated into the weighted cell–gene graph. For each node (i.e. cell-centered and gene-centered) subgraph, weighted edges were normalized and the node itself and its neighborhood were then gathered to generate a new cell node representation during aggregation. The linear classifier layer categorizes the final cell state representation as a predefined cell type.CC BY-NC · PubMed Central
Figure 3. Performance and robustness of scDeepSort on internal test datasets and performance of different references on annotation of cell types. (A) Accuracy of scDeepSort in annotating the HCL test set across 56 tissues (line chart) and the MCA test set across 32 tissues with five replicates by the 8:2 split of the training set and the test set. The bar chart shows the number of test cells for each tissue. (B) Accuracy of scDeepSort in annotating the HCL and MCA test set by the split of 8:2, 7:3, 6:4 and 5:5 with five replicates for each alternative split. The data of the accuracy represents the mean ± SD. (C) Performance of different references on the annotation of cell types. The accuracy for each dataset was labelled for bar plot (left) and the median accuracy across seven datasets was labelled beside the box (right).
Figure 3. Performance and robustness of scDeepSort on internal test datasets and performance of different references on annotation of cell types. (A) Accuracy of scDeepSort in annotating the HCL test set across 56 tissues (line chart) and the MCA test set across 32 tissues with five replicates by the 8:2 split of the training set and the test set. The bar chart shows the number of test cells for each tissue. (B) Accuracy of scDeepSort in annotating the HCL and MCA test set by the split of 8:2, 7:3, 6:4 and 5:5 with five replicates for each alternative split. The data of the accuracy represents the mean ± SD. (C) Performance of different references on the annotation of cell types. The accuracy for each dataset was labelled for bar plot (left) and the median accuracy across seven datasets was labelled beside the box (right).CC BY-NC · PubMed Central
Figure 4. Performance comparison on human external test datasets. (A) Heatmaps of accuracies for scDeepSort, CellAssign, Garnett, SingleR, scMap-cell, scMap-cluster, ACTINN, CHETAH, scID, SCINA, scPred, singleCellNet and SVM on 27 external test datasets across 10 tissues and mean F1 scores and mean MCCs on 20 external test datasets across nine tissues. The asterisk represents the top-ranked method for each test dataset. NA, not available. (B) Boxplots summarized the maximal, minimal, median and quantile accuracies, mean F1 scores and mean MCCs for different methods among different tissues. The median value is labelled beside the corresponding box. MCC, Matthews correlation coefficient.
Figure 4. Performance comparison on human external test datasets. (A) Heatmaps of accuracies for scDeepSort, CellAssign, Garnett, SingleR, scMap-cell, scMap-cluster, ACTINN, CHETAH, scID, SCINA, scPred, singleCellNet and SVM on 27 external test datasets across 10 tissues and mean F1 scores and mean MCCs on 20 external test datasets across nine tissues. The asterisk represents the top-ranked method for each test dataset. NA, not available. (B) Boxplots summarized the maximal, minimal, median and quantile accuracies, mean F1 scores and mean MCCs for different methods among different tissues. The median value is labelled beside the corresponding box. MCC, Matthews correlation coefficient.CC BY-NC · PubMed Central
Figure 5. Performance comparison on mouse external test datasets. (A) Heatmaps of accuracies for scDeepSort, CellAssign, Garnett, SingleR, scMap-cell, scMap-cluster, ACTINN, CHETAH, scID, SCINA, scPred, singleCellNet and SVM on 49 external test datasets across 18 tissues and mean F1 scores and mean MCCs on 11 external test datasets across six tissues. The asterisk represents the top-ranked method for each test dataset. NA, not available. (B) Boxplots summarized the maximal, minimal, median and quantile accuracies, mean F1 scores and mean MCCs for different methods among different tissues. The median value is labelled beside the corresponding box. MCC, Matthews correlation coefficient.
Figure 5. Performance comparison on mouse external test datasets. (A) Heatmaps of accuracies for scDeepSort, CellAssign, Garnett, SingleR, scMap-cell, scMap-cluster, ACTINN, CHETAH, scID, SCINA, scPred, singleCellNet and SVM on 49 external test datasets across 18 tissues and mean F1 scores and mean MCCs on 11 external test datasets across six tissues. The asterisk represents the top-ranked method for each test dataset. NA, not available. (B) Boxplots summarized the maximal, minimal, median and quantile accuracies, mean F1 scores and mean MCCs for different methods among different tissues. The median value is labelled beside the corresponding box. MCC, Matthews correlation coefficient.CC BY-NC · PubMed Central

Verification

Claims the paper makes

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

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

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 66 references

Cited by

See all 158

Metadata

Authors
Shao, X., Yang, H., Zhuang, X., Liao, J., Yang, P., Cheng, J., Lu, X., Chen, H., & Fan, X.
Affiliations
  • Zhejiang University (CN)
  • Pharmaceutical Informatics Institute, College of Pharmaceutical Sciences, Zhejiang University, Hangzhou 310058, China
  • First Affiliated Hospital Zhejiang University (CN)
  • Westlake University (CN)
  • The University of Sydney (AU)
Corresponding author
  • Chen, Huajun
  • Fan, Xiaohui
Venue
Nucleic Acids Research
Published
2021-09-09
Volume
49
Issue
21
Pages
e122-e122
ISSN
  • 0305-1048 (ISSN-L)
  • 1362-4962
OpenAlex
W3198628780
Cited by
158
References
66
Citations by year
2026: 23 · 2025: 35 · 2024: 47 · 2023: 31 · 2022: 20 · 2021: 2
FWCI
7.88
Type
journal article
Language
en
Open access
Locations
  • Nucleic Acids Research (publishedVersion, cc-by-nc)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
Funders
  • National Natural Science Foundation of China (81973701, 91846204/U19B2027)
  • Natural Science Foundation of Zhejiang Province (LZ20H290002)
  • National Youth Top-notch Talent Support Program (W02070098)
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 expression and cancer classification
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Phylogenetic Studies
Keywords
cell type annotation, single-cell transcriptomics, pre-trained deep learning models, human and mouse tissues, cross-dataset generalization, single-cell RNA sequencing
MeSH terms
Deep Learning, Animals, Humans, RNA, Neural Networks, Computer, Databases, Genetic, Mice, Single-Cell Analysis, Transcriptome

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: scDeepSort: a pre-trained cell-type annotation method for single-cell transcriptomics using deep learning with a weighted graph neural network. https://app.cochat.ai/papers/10.1093/nar/gkab775
The paper
Shao, X., Yang, H., Zhuang, X., Liao, J., Yang, P., Cheng, J., Lu, X., Chen, H., & Fan, X. (2021). scDeepSort: a pre-trained cell-type annotation method for single-cell transcriptomics using deep learning with a weighted graph neural network. Nucleic Acids Research. https://doi.org/10.1093/nar/gkab775

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.