ezSingleCell: An integrated one-stop single-cell and spatial omics analysis platform for bench scientists

doi:10.21203/rs.3.rs-2699698/v1

Download PDF

Article

ezSingleCell: An integrated one-stop single-cell and spatial omics analysis platform for bench scientists

https://doi.org/10.21203/rs.3.rs-2699698/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 03 Jul, 2024

Read the published version in Nature Communications →

Version 1

posted

You are reading this latest preprint version

ezSingleCell is an interactive and easy-to-use application for the analysis and integration of multiple single-cell and spatial omics data types. It integrates the best-performing publicly available methods and in-house novel algorithms for in-depth data analysis, integration, and interactive data visualization. By integrating the relevant tools to form a complete analysis pipeline, we save users from the hassles of choosing among the enormous diversity of available methods, installing individual packages, and reformatting the data for different analysis steps. ezSingleCell takes input data in a variety of formats such as text files or Cell Ranger/ Space Ranger output and produces publication ready figures and tables. Users can customize the relevant parameters to ensure the quality and accuracy of their data analysis. Users can also download and store the R objects from ezSingleCell to perform additional offline analyses. ezSingleCell’s streamlined interface can analyze a standard scRNA-seq dataset containing 3000 cells in less than five mins. ezSingleCell is also accompanied by an in-depth manual and video tutorials to guide users. Overall, these features make ezSingleCell a convenient and easy web service for single cell and spatial analysis without requiring prior programming knowledge. ezSingleCell is available in two forms: an installation-free web application (https://immunesinglecell.org/ezsc/) or a software package with a shinyApp interface (https://github.com/JinmiaoChenLab/ezSingleCell2) that runs on a personal computer with a low memory requirement of 8 Gb RAM.

Biological sciences/Computational biology and bioinformatics/Software

Biological sciences/Computational biology and bioinformatics/Programming language

Single-cell RNA sequencing (scRNA-seq) has emerged as a powerful technology to acquire gene expression profiles at the single-cell level¹. This enables novel insights into the heterogeneity of biological systems². Single-cell measurements has also been extended to other omics such as scATAC-seq and CITE-seq³ to measure chromatin accessibility and proteins, respectively. In addition, advances in spatial transcriptomics are now enabling researchers to probe tissue samples at single cell resolution while retaining their spatial context^4–7. The latter feature is crucial in understanding how the different cell types are spatially arranged to give rise to the tissue’s emergent properties.

The rapid rise of single-cell technologies has led to enormous amounts of data being generated. New tools are also being developed to exploit the data generated. In 2021, Zappia L and Theis FJ reported that the number of catalogued single-cell tools in the scRNA-tools database exceeded 1000⁸. Currently, two software platforms remain dominant; Seurat⁹ and Scanpy¹⁰ are the de facto standards for single cell analysis in R and Python, respectively. Both require a minimum level of bioinformatics expertise and coding knowledge, thereby creating a barrier to data analysis for bench scientists. In contrast, tools with intuitive graphic user interfaces would greatly benefit the single cell community as a whole. There are several recently developed web servers from both the research community and commercial companies (Table 1), but the majority are limited to transcriptomics analysis only and cannot handle spatial omics, single-cell multi-omics, and single-cell chromatin accessibility (scATAC-Seq)-derived data in a unified interface. Moreover, these tools mainly provide basic analysis functionalities. More advanced down-stream analyses such as cell type identification and cell-cell interaction are not included in their pipelines. Only ICARUS¹¹ and CELLAR¹² can handle multiple modalities and provide some downstream analysis such as cell-type identification. However, ICARUS and CELLAR do not offer other downstream analysis such as cell-cell communication and cell type deconvolution from spatial datasets. Another webserver, SciAp¹³ (https://humancellatlas.usegalaxy.eu/), integrates tools from different workflows, including 20 modules from Scanpy, covering data filtering, normalization, variable gene selection, clustering, dimensionality reduction, and trajectory inference methods, but currently handles scRNA-seq data only. Moreover, integrative analysis of different data modalities, such as integrating scRNA-seq with spatial transcriptomics, is often desired by scientists to obtain a holistic view of the biological systems under study. Such analyses are not possible with current web servers.

To provide a more comprehensive data analysis platform with a user-friendly interface, we present ezSingleCell, an integrated one-stop single-cell and spatial analysis web server for bench scientists (https://immunesinglecell.org/ezsc/). ezSingleCell accepts data input in multiple formats such as text, csv, tsv, or 10X Cell Ranger/Space Ranger output, and returns publication ready figures and tables. ezSingleCell improves on existing single cell data analysis web servers including SciAp, ICARUS, and CELLAR in the following aspects. Firstly, ezSingleCell encompasses a much larger scope of single cell data analysis, including single cell multi-omics and spatial transcriptomics. The tools offered for these analyses include our in-house algorithms, GraphST¹⁴ and CELLiD¹⁵. Secondly, ezSingleCell provides many advanced downstream analysis capabilities such as cell-cell communication and cell type deconvolution of spatial datasets. Finally, scATAC-seq datasets can be analysed with ezSingleCell, a functionality that most webserver lacks.

Overview of ezSingleCell and its advantages over other tools

ezSingleCell consists of five modules, namely the single-cell RNA-Seq (scRNA-Seq) module, single-cell data integration (scIntegration) module, spatial transcriptomics (ST) module, single-cell multiomics (scMultiomics) module and single-cell ATAC-seq (scATAC-Seq) module (Fig. 1). Each module offers multiple tools for complete data analysis workflows, starting from data pre-processing to interactive results visualization. For each analysis step, we selected several top performing methods based on our and others’ benchmarking studies. Both published and in-house novel algorithms are available in ezSingleCell as listed in Table 4. For each analysis step, default parameters are provided to achieve optimal results for most cases. Users can also modify them to suit their data. In the scRNA-Seq analysis module, the user can perform basic analyses such as clustering and differential gene expression analysis, as well as more advanced analyses including cell type identification, gene set enrichment analysis, and predicting cell-cell communication. The scIntegration module offers the top three performing data integration algorithms for single-cell transcriptomics, namely Seurat Integration, Harmony and fastMNN. We also included several quantitative metrics for the users to assess batch integration performance. The ST module is equipped with the Seurat spatial transcriptomics analysis functionalities and our in-house GraphST algorithm. These tools enable spatial clustering, multi-sample integration, and cell type deconvolution of spatial data acquired with different technology platforms such as Visium and sub-cellular technologies like Xenium. In the scMultiomics module, we evaluated all available methods and selected Seurat WNN and MOFA + to be included for the multiomics data analysis. These methods can handle data including CITE-Seq (joint scRNA-Seq with protein), Multiome (joint scRNA-Seq with ATAC-Seq). Finally, in the scATAC-Seq module, Signac algorithm is employed as it provides end-to-end analysis of single-cell chromatin accessibility data, including peak calling, quantification, quality control, dimension reduction, clustering, integration with single-cell gene expression datasets, DNA motif analysis, and interactive visualization.

Here we extensively compare the features of ezSingleCell with publicly available webservers and commercial services. As shown in Tables 2 and 3, most of the web servers (such as SciAp, ASAP, alona, NASQAR, SCTK, Asc-seurat) are limited to transcriptomics analysis only. Some web services do provide additional analysis such as data integration (SCTK 2.0, ICARUS), scMultiomics (ICARUS, CELLAR), scATAC-seq (shinyArchR.UiO), and spatial analysis (CELLAR) but they do not offer a unified interface for comprehensive analysis. ezSingleCell also provides advanced downstream analysis functionalities such as cell-cell communication and cell type deconvolution from spatial datasets. Moreover, ezSingleCell provides support for scATAC-seq dataset analysis which most webservers lack.

ezSingleCell is available in two forms: an installation-free web application (https://immunesinglecell.org/ezsc/), or a software package with a Shiny app interface (https://github.com/JinmiaoChenLab/ezSingleCell2) that can be run on a computer with basic memory requirements (8 Gb RAM).

ezSingleCell’s scRNA-seq module streamlines the analysis of scRNA-seq data

The scRNA-seq analysis module of ezSingleCell relies on Seurat for basic analysis and other packages / in-house algorithms for advanced analyses such as cell type annotation, gene set enrichment analysis (GSEA) and cell-cell communication. Here we illustrate the utility of ezSingleCell for scRNA-seq analysis (Figs. 2 and 3). As an example, we used the dataset of 2700 peripheral blood mononuclear cells (PBMCs) from the Seurat guided clustering tutorial vignette (Table 5). ezSingleCell allows users to perform quality control and filter out low quality cells using parameters such as min.genes and min.cells, and visualize the data using violin plots, feature plots, and ridge plots (Supplementary Fig. 3A). Users can then perform data pre-processing using either log-normalization or SCTransform and select the desired number of variable features. In this example, we used the default parameters of min.genes = 200 and min.cells = 3. We then performed data pre-processing with log-normalization, variable feature selection for 2000 genes, and data scaling. This was followed by Principal Component Analysis (PCA). Cell clustering was then performed using the first 10 PC dimensions, with a k-nearest neighbour value of 10 and Louvain community detection with a clustering resolution of = 2 to obtain 20 clusters. The results were visualized using UMAP at dims = 1:10 (Fig. 2B).

We next annotated the clusters with our in-house cell identification algorithm, CELLiD (https://www.immunesinglecell.org/cellpredictor) using the ‘blood’ cell type reference. Using CELLiD, we identified 15 cell types: Memory CD4 T, Naive CD8 T, CD14 monocyte, Memory B, Naive CD4 T, Naive B, CD16 monocyte, GZMB CD8 T, CD16 NK, GZMK CD8 T, MAIT, CD14/CD16 monocyte, cDC, Platelet and Proliferation T/NK (Fig. 2C and supplementary Fig. 3B). We were able to recapitulate the original annotations provided by Seurat vignette except that we annotated one portion of the Naïve CD4 T cells as Naïve CD8 and a subset of CD8 T as MAIT cells. We confirmed the correctness of our annotation by checking marker gene expression (supplementary Fig. 3F). Furthermore, we could split the B cells into Naïve and Memory B subsets, CD8 T cells into GZMK and GZMB CD8 T cell subsets, providing a higher resolution of cell type labelling. We then identified differentially expressed genes, such as GZMK and CCL5 for MAIT cells, CD79A, CD79B, and MS4A1 for the B cell subsets, and S100A8, S100A9 and LYZ for CD14 monocytes (Fig. 2D). Visualization of these genes was also done using violin plots, feature plots, and ridge plots which confirmed their higher expression in these cell type clusters (Supplementary Fig. 3C, 3D, and 3E). Users can also identify DEGs between two specified cell types and visualize the DEGs using a volcano plot where the significant top genes with the highest differential expression were highlighted (Fig. 2E). For example, we observed FCGR3A, S100A8, and LYZ to be differentially upregulated in CD16 monocytes as compared to CD14 monocytes.

We then computed cell type similarity (Fig. 3A), which showed consistency with the clusters observed in the UMAP plot. The cell type in a large cluster showed higher similarity score with other cell types in the same cluster, whereas isolated clusters showed low similarity scores with other cell type clusters. This analysis of cell type similarity is useful when users identify a population of unknown cells and wish to estimate its similarity to known cell types.

We also offer the fgsea function¹⁶ for users to determine whether an a priori defined gene set is differentially enriched between two cell types or states (Fig. 3B). Users can choose any gene set from the MSigDB database for human (Hallmark, C1-C8) and mouse (MH, M1, M2, M3, M5 and M8). Here we performed GSEA analysis between the Naïve CD4 T and Memory CD4 T cells using the Human ‘C7’ immunologic signature gene set. We ranked the genes based on their log fold change and identified the top enriched pathways. As expected, we observed GSE11057 NAÏVE VS MEMORY CD4 TCELL UP, GSE11057 NAIVE VS EFF MEMORY CD4 TCELL UP and GSE11057 NAIVE VS CENT MEMORY CD4 TCELL UP were over-represented in Naïve CD4 compared to Memory CD4. Other gene sets that also showed significant differences are listed in Fig. 3B. Lastly, ezSingleCell incorporates the widely adopted CellPhoneDB package and various resource databases of ligand-receptor pairs for analysing cell-cell communication. We used CellPhoneDB to predict cell-cell communications and found that cDC and CD16 monocyte shows the highest number of interactions (Fig. 3C top). We then selected the top significant ligand-receptor pairs between cDCs and CD16 monocytes using a p-value cut-off < 0.05 (Fig. 3C bottom). These agree with the gene expression patterns for genes such as LILRB4, GRN, LGALS9, C5AR1 and TNFSF13B (Supplementary Fig. 3G).

ezSingleCell’s scIntegration module performs batch correction of multiple scRNA-seq datasets

To demonstrate batch effect correction of scRNA-seq data in ezSingleCell, we used a human PBMC dataset comprised of 2 batches. Both datasets were obtained from 10x Genomics, with the 3’ batch data consisting of 8381 and the 5’ batch data containing 7726 cells (Table 5). The cell type annotations were retrieved from previous studies^17,18. For batch effect correction, ezSingleCell takes in both the expression datasets and the metadata with batch information. Cell type information is optional but required for cell type separation assessment downstream. We first performed the standard quality control and data pre-processing steps with log-normalization, finding the top 2000 highly variable genes, and scaling using the default parameters. Clustering was done with the first 10 PCs, k-nearest neighbour value of 10, and Louvain clustering at a resolution of 0.6. In the UMAP plot before batch correction, we can easily observe the batch effects present (Fig. 4). Three batch correction methods are currently available in ezSingleCell, Seurat, Harmony, and fastMNN. We ran the 3 methods with 2000 integration features and performed PCA, clustering, and UMAP on their results. We note that Seurat, Harmony, and fastMNN were able to successfully remove the batch effects (Fig. 4). To benchmark the batch mixing, we computed the median iLISI scores in ezSingleCell to assess batch mixing with a higher value denoting better performance. If the batches are well-mixed, a maximum score of 2 (matching the number of data batches) can be achieved. Harmony ranked first, followed by fastMNN and Seurat, achieving iLISI scores of 1.70, 1.54 and 1.43, respectively (Fig. 4). Following batch integration, other downstream analysis such as cell type identification using the CELLiD algorithm, differential gene expression analysis, and marker gene visualization can be performed (Supplementary Fig. 4).

ezSingleCell’s ST module performs clustering, integration, and deconvolution of spatial transcriptomics

In this module, we incorporated Seurat and GraphST for processing and analysis of spatial transcriptomics data. Seurat offers functionality for both data pre-processing and clustering. GraphST, a state-of-the-art graph self-supervised contrastive learning method that incorporates spatial location information and gene expression profiles, can achieve best in class performance for spatial clustering, multi-sample integration, and cell type deconvolution. The module can be used to analyse data acquired from different platforms such as 10x Visium and Stereo-seq that offers sub-cellular solution. In this demonstration, we analysed a mouse brain sagittal anterior dataset acquired with 10x Visium and processed with the Space Ranger pipeline v1.1.0 (Table 5). We first performed quality control, followed by the pre-processing steps of normalization using SCTransform, selection of 2,000 highly variable genes, and data scaling. Users can first inspect the gene expression values on the tissue slide (Fig. 5A). Next, we performed PCA using default parameters. For spatial clustering, we applied both Seurat and GraphST. We used the first 10 PCs, KNN value of 10, and Louvain clustering resolution of 0.6, followed by dimension reduction with UMAP to visualize the 15 clusters found (Fig. 5B). We also set the ‘n_cluster’ = 15 for GraphST and observed that the clustering from GraphST was more consistent with the manual annotation (Fig. 5B). We then assessed the spatial clustering using the mean silhouette score metric.

ezSingleCell also offers the capability to deconvolute or annotate cell types in spatial spots using an annotated single cell reference. For this purpose, scRNA-seq reference data is first loaded and processed using the standard pre-processing pipeline. We can then execute the deconvolution using the ‘label transfer’ function from Seurat or ‘project_cell_to_spot’ function from GraphST. The results can be interactively visualized and showed (Fig. 5C).

ezSingleCell also offers analysis support for sub-cellular solution data such as data acquired using the Xenium platform. Users can perform clustering analysis and interactively visualize the expression patterns at sub-cellular level. In addition, users can zoom into a certain section to have a closer look at the cellular composition and inter-cellular interactions. Users can also view the expression profile of each gene on the tissue slice (Fig. 5D).

When tissue samples are larger than the capture slides used for spatial transcriptomics, horizontal integration enables the data from multiple capture slides to be stitched together. Here we show GraphST’s horizontal integration capabilities using the mouse brain 10x Visium data with two sections, both divided into anterior and posterior slices. We first aligned the tissue slices of each section and performed clustering. Compared to Seurat’s output, GraphST clusters showed better agreement with known anatomy. In particular, we note that GraphST’s clusters were more accurate in capturing the structures of the cerebral cortex layers, cerebellum, and corpus callosum. This highlights GraphST’s consistent capability in aligning shared domains across tissue sections and meanwhile removing batch effects (Supplementary Fig. 5D).

ezSingleCell’s scMultiomics module performs joint analysis of multiple modalities

We next demonstrate ezSingleCell for multi-modal single cell analysis with healthy PBMC datasets. The datasets consist of gene expression and protein expression (CITE-seq) data, as well as gene expression and chromatin accessibility. Both datasets were downloaded from 10x Genomics (Table 5), where the CITE-seq data has 7865 cells and the second dataset (MultiOme dataset) comprised of 3012 cells. The first dataset was pre-processed with the default parameters and the second dataset with SCTransform normalization. Cell clustering was then performed using the first 10 PC dimensions, k-nearest neighbour value of 10, and clustering resolution of 0.6 (Fig. 6A).

For joint modality analysis of the CITE-seq data, Seurat WNN, a weighted combination of RNA and protein similarities, was used to calculate the KNN graph for clustering. This yielded 20 clusters for both individual modality and joint analysis. With CELLiD, we identified 16 cell types at a resolution of 1 (Fig. 6A). We next visualized the relevant markers such as MS4A1 for B cells and CD4-TotalSeqB in both RNA and ADT assays respectively to verify the cell type annotations (Fig. 6A). We can also use DEG analysis to further verify the cell types by checking the top differentially expressed genes of each cluster. We ran one more method, i.e. MOFA + to perform multimodal analysis and cluster the data by specifying the number of clusters (or k) = 10 and using all latent factors computed using MOFA. Using CELLiD, we identified 10 celltypes. Then, we visualized the marker genes such as MS4A1 for B cells and CD4-TotalSeqB in both RNA and ADT assays respectively to verify the cell type annotations (Fig. 6A).

ezSingleCell’s scATAC-Seq module performs scATAC-seq data analysis and integration

In this final demonstration, we employed ezSingleCell to illustrate ATAC-seq processing using a dataset of 500 PBMCs from a healthy donor. The output from Cell Ranger ATAC (Table 5) was downloaded from 10x Genomics. For quality control, the user can compute metrics such as nucleosome banding pattern, transcriptional start site (TSS) enrichment score, total number of fragments in peaks, and ratio reads in genomic blacklist regions (Fig. 7B). We normalized the data using Term Frequency-Inverse Document Frequency (TF-IDF) normalization, followed by feature selection and dimension reduction. Cell clustering was then performed using the first 10 PC dimensions with a k-nearest neighbour value of 10 and a Louvain clustering resolution of 0.6. The UMAP was then computed to visualize the 4 clusters found (Fig. 7C). We next performed differential peak analysis to find differentially expressed peaks between clusters. These were then visualized using violin, feature, and coverage plots (Fig. 7D). ezSingleCell also offers integration of scATAC-seq with scRNA-seq data to accomplish label transfer of cell type information. In this example, we could identify 12 cell types in the clusters (Fig. 7F).

We next performed motif analysis using Signac and it revealed potentially important cell-type-specific regulatory sequences such as NRF1, ZBED1, ZBTB33, TCFL5, YY2, and BARHL1 (Fig. 7E). Transcription factor foot printing was then visualized for these sequences (Fig. 7G). We computed the pairwise co-accessibility scores for each peak and grouped the pairwise connections into larger co-accessible networks using the generate_ccans function from Cicero (Fig. 7H). The results are visualized with a coverage plot (Fig. 7D).

We can also employ ezSingleCell to integrate multiple scATAC-Seq datasets. Here we integrated two scATAC-seq datasets (Table 5) [Figure 7I] and assessed the batch correction efficacy with the LISI metric. With this dataset, we obtained a score of 1.77 indicating good batch mixing (a maximum score of 2 matching the number of batches indicates that the batches are completely mixed).

Single-cell profiling of cells is generating huge amounts of data. Despite the increasing availability of analytical tools to exploit the data, data analysis by bench scientists is hampered by minimum bioinformatics skill requirements of these tools. ezSingleCell is an integrated one-stop single-cell and spatial omics analysis platform with an intuitive GUI suited for users with no bioinformatics background. ezSingleCell has analysis modules to cover the data generated by different single-cell omics experiments, including scRNA-seq, scIntegration, scMultiomics (CITE-seq, 10x Multiome), scATAC-Seq, and spatial omics. ezSingleCell accomplishes this by combining in-house novel algorithms such as CELLiD for cell type identification and GraphST for spatial clustering and deconvolution, along with other established methods for both basic and advanced analyses such as batch effect correction, gene set enrichment analysis, cell-cell communication, and spatial deconvolution. The GUI is designed to be user-friendly for interactive data exploration and analysis. Users can customize their data analysis with a wide range of parameters. ezSingleCell also accepts inputs in different formats such as text files or Cell Ranger / Space Ranger output and returns publication ready figures and tables. ezSingleCell is available in two forms: an installation-free web application (https://immunesinglecell.org/ezsc/), or a software package with a Shiny app interface (https://github.com/JinmiaoChenLab/ezSingleCell2) that can be run on a computer with basic memory requirements (8 Gb RAM).

We will continue to maintain and upgrade ezSingleCell. As the number of cells profiled per experiment is increasing rapidly, this results in very large datasets being generated and consequently analysis and integration of such big data is time consuming and memory intensive. In future versions of ezSingleCell, we will incorporate novel deep learning methods for efficient dimension reduction, clustering, and batch integration. Moreover, we will enable cross talk between the five ezSingleCell modules so that different data modalities can be analyzed jointly. We will also extend the current spatial transcriptomics module to analyze spatial proteomics and spatial multi-omics. We will also enable inference of cell-cell interactions based on spatial proximity and the expression of ligand-receptor pairs. With the emergence of the latest sub-cellular spatial technologies such as Nanostring CosMX, StereoSeq, Vizgen MERSCOPE, PixelSeq, SeqScope, we will incorporate novel algorithms for better cell segmentation to process such datasets.

Data input

ezSingleCell accepts input from multiple technologies (Smart-Seq2, 10x, CITE-Seq, Multiome, Visium) in a variety of formats: (i) text files (txt, csv, or tsv) or 10x Cell Ranger output for scRNA-seq and data integration analysis, (ii) 10x CITE-Seq count output for CITE-Seq, and 10x Cell Ranger ARC output for multiome, (iii) 10x Cell Ranger ATAC output for scATAC-Seq analysis, and (iv) 10X Space Ranger output for spatial transcriptomics analysis.

Data pre-processing

ezSingleCell offers quality control and data normalization with log-normalization or SCTransform¹⁹. The normalized data can then be scaled, and Principal Component Analysis (PCA) is performed. For sc-multi-omics data, ezSingleCell has 3 different methods to perform multimodal analysis, namely Seurat WNN⁹ and MOFA + ²⁰. For Seurat WNN, both assays (RNA and ADT for CITE-seq data; RNA and ATAC for 10x scMultiOme data) are first pre-processed and dimensional reduction performed independently. The closest neighbours are thereafter calculated for each cell based on a weighted combination of RNA and protein similarities. The resulting WNN is usable for visualization and clustering. After pre-processing, MOFA + infers a low-dimensional representation of the data in terms of a small number of (latent) factors that capture the global sources of variability. MOFA + employs Automatic Relevance Determination (ARD), a hierarchical prior structure that facilitates untangling variation that is shared across multiple modalities from variability that is present in a single modality.

For scATAC-Seq data, ezSingleCell employs the Signac package²¹ to perform term frequency-inverse document frequency (TF-IDF) normalization. This is a two-step normalization procedure, that both normalizes across cells to correct for differences in cellular sequencing depth, and across peaks to give higher values to more rare peaks. This is followed by feature selection using only the top n% of features (peaks) for dimensional reduction or removing features present in less than a specified number of cells. Singular value decomposition (SVD) is then performed on the TD-IDF matrix of the selected features (peaks) to return a reduced dimension representation. This workflow is known as latent semantic indexing (LSI). Users can visualize the correlation between each LSI component and sequencing depth.

For spatial transcriptomics data, quality control functions are available and data normalization is done using SCTransform²². In spatial data, the variance in molecular counts per spot can be substantial and is also dependent on the tissue anatomy, particularly if there are differences in cell density across the tissue. For example, tissues that are depleted for neurons (such as the cortical white matter), reproducibly exhibit lower molecular counts. As a result, standard approaches such as log normalization which forces all cells or spots to have the same total count, are inappropriate. Therefore, SCTransform is recommended over log-normalization as it builds on regularized negative binomial models of gene expression to account for technical artifacts while preserving biological variance.

Clustering and dimension reduction

Clustering in ezSingleCell employs graph-based community detection where a k nearest neighbour (KNN) graph is constructed based on the Euclidean distance in the PCA space and the edge weights between two cells are refined based on the shared overlap in their local neighbourhoods (Jaccard similarity). Modularity optimization techniques such as the Louvain algorithm²³ are then applied to iteratively cluster the cells together to optimize the standard modularity function. The user can adjust the clustering resolution to obtain broad or fine-grained clusters with increased values leading to a greater number of clusters. For dimensionality reduction, UMAP and tSNE are available. The user can specify the number of PCA dimensions to be used.

For scMulti-omics data, closest neighbours of each cell are calculated based on a weighted combination of RNA and protein similarities. The cell-specific modality weights and multimodal neighbours are calculated using a function ‘FindMultiModalNeighbors’. Clustering is then performed using a shared nearest neighbour (SNN) modularity optimization technique such as the Louvain algorithm.

To cluster spatial data, ezSingleCell offers two methods, namely Seurat and GraphST. Seurat use a graph-based community detection algorithm such as Louvain clustering where a KNN graph is constructed based on the Euclidean distance in the PCA space. GraphST combines graph neural networks with self-supervised contrastive learning to learn latent representations of spots from gene expression and spatial information to accomplish spatial clustering.

Cell type identification

To identify cell types of clusters, ezSingleCell implements our in-house CELLiD algorithm. CELLiD employs the average expression of clusters for correlation analysis with the reference dataset from DISCO database (https://www.immunesinglecell.org/) to infer the cell type. It is available in the Single cell RNA-seq, scIntegration, and scMultiomics modules.

Data visualization

ezSingleCell offers UMAP, tSNE, violin, feature, ridge, and volcano plots for data visualization. For spatial datasets, users can visualize their data with interactive plots showing spatial localization. For scATAC-seq datasets, users can visualize their data using CoveragePlot showing the transcription factors and peaks within given regions of the genome.

Batch effect correction and data integration

ezSingleCell contains four popular and best-performing methods namely Seurat, Harmony²⁵ and fastMNN²⁶ identified from previous benchmarking studies^18,27. In this data integration step, users can customise tuning parameters such as number of features to integrate. Users can quantitatively benchmark the data integration results with metrics such as ARI, iLISI, and cLISI. For scATAC-seq datasets, batch effect removal in ezSingleCell uses the reciprocal LSI projection to find integration anchors between the two datasets, followed by integration of the datasets in the low-dimensional cell embeddings (the LSI coordinates). For spatial datasets, GraphST enables vertical and horizontal batch integration of multiple tissue sections.

scRNA-seq downstream analyses

Differentially expressed genes analysis is available in ezSingleCell using the ‘FindAllMarkers’ function from the Seurat package with statistical tests like Wilcoxon, bimod, and t-test available. Available downstream analyses include gene set enrichment analysis (GSEA)²⁸ analysis with the msigdbr database, and cell-cell communication with the CellphoneDB²⁹ package. Users can select from multiple reference resources such as ICELLNET, CellPhoneDB, CellChatDB, CellTalkDB and, and OmniPath for cell-cell communication prediction.

Single cell ATAC-seq module

For processing ATAC-Seq data, ezSingleCell uses MACS2 to perform peak calling. Users can visualize the results with Coverage Plot and perform differential accessibility (DA) test to find differentially accessible regions between clusters using logistic regression. ezSingleCell is also equipped with chromVAR for analysing chromatin accessibility variations in scATAC-Seq data. It computes the gain or loss of chromatin accessibility for known transcription factor motifs or de novo sequence motifs. It is computed on a per-cell basis and can be visualized for different motif sequences. Differentially active motifs can be further computed between different cell clusters.

ezSingleCell also offers the ability to merge multiple scATAC-Seq datasets by performing peak calling on each dataset independently and create a common set of peaks across all datasets to be merged. The combined set of peaks in each dataset is then quantified and the shared set of features across datasets are merged.

Integrating scRNA-seq and scATAC-seq data

ezSingleCell allows users to integrate their scATAC-seq data with scRNA-seq data by performing cross-modality integration and label transfer. In the integration process, ezSingleCell estimates the RNA-seq levels from ATAC-seq by quantifying gene expression ‘activity’ from ATAC-seq reads. The internal structure of the ATAC-seq data is then learned using LSI. The ‘anchors’ are then identified between the ATAC-seq and RNA-seq datasets followed by data transfer between datasets (either transfer labels for classification or impute RNA levels in the ATAC-seq data to enable co-embedding). Finally, the scRNA-seq and scATAC-seq cells are co-embed in the same low dimensional space for visualization using the same anchors to transfer cell type labels to impute scRNA-seq values for the scATAC-seq cells. The measured and imputed scRNA-seq data are then merged and can be visualized via a UMAP.

Integration of scRNA-seq and spatial data (deconvolution)

Users can employ two different methods, namely Seurat and GraphST to integrate their spatial data with scRNA-seq reference data to annotate the cell types present. Publicly available scRNA-seq references are available at the DISCO database (https://www.immunesinglecell.org/) or the Human Cell Atlas (https://www.singlecell.broadinstitute.org) for human data, and Allen Brain Atlas data portal (http://www.brain-map.org) for mouse brain data.

DATA AVAILABILITY

The data used in this study has been uploaded to Github and is freely available at:

https://github.com/JinmiaoChenLab/ezSingleCell2.

SUPPLEMENTARY DATA

Supplementary Data are available at Nature Methods online.

ACKNOWLEDGEMENT

I would like to acknowledge members of Jinmiao’s lab in SigN, A*STAR.

FUNDING

The work was supported by A * STAR, AI, Analytics and Informatics (AI3) Horizontal Technology Programme Office (HTPO) seed grant C211118015, the Singapore National Medical Research Council Open Fund Individual Research Grant [NMRC-OFIRG18nov-2013], Singapore National Medical Research Council Open Fund Large Collaborative Grant [NMRC/OFLCG/003/2018] and the National Research Foundation (NRF-CRP19-2017-04).

CONFLICT OF INTEREST

The authors declare no conflict of interest.

Chen, G., Ning, B. & Shi, T. Single-cell RNA-seq technologies and related computational data analysis. Frontiers in Genetics vol. 10 Preprint at https://doi.org/10.3389/fgene.2019.00317 (2019).
Haque, A., Engel, J., Teichmann, S. A. & Lönnberg, T. A practical guide to single-cell RNA-sequencing for biomedical research and clinical applications. Genome Medicine vol. 9 Preprint at https://doi.org/10.1186/s13073-017-0467-4 (2017).
Stoeckius, M. et al. Simultaneous epitope and transcriptome measurement in single cells. Nat Methods 14, 865–868 (2017).
Marx, V. Method of the Year: spatially resolved transcriptomics. Nat Methods 18, 9–14 (2021).
Moses, L. & Pachter, L. Museum of spatial transcriptomics. Nature Methods vol. 19 534–546 Preprint at https://doi.org/10.1038/s41592-022-01409-2 (2022).
Rao, A., Barkley, D., França, G. S. & Yanai, I. Exploring tissue architecture using spatial transcriptomics. Nature vol. 596 211–220 Preprint at https://doi.org/10.1038/s41586-021-03634-9 (2021).
Williams, C. G., Lee, H. J., Asatsuma, T., Vento-Tormo, R. & Haque, A. An introduction to spatial transcriptomics for biomedical research. Genome Medicine vol. 14 Preprint at https://doi.org/10.1186/s13073-022-01075-1 (2022).
Zappia, L. & Theis, F. J. Over 1000 tools reveal trends in the single-cell RNA-seq analysis landscape. Genome Biology vol. 22 Preprint at https://doi.org/10.1186/s13059-021-02519-4 (2021).
Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573-3587.e29 (2021).
Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: Large-scale single-cell gene expression data analysis. Genome Biol 19, (2018).
Jiang, A., Lehnert, K., You, L. & Snell, R. G. ICARUS, an interactive web server for single cell RNA-seq analysis. Nucleic Acids Res 50, W427–W433 (2022).
Hasanaj, E., Wang, J., Sarathi, A., Ding, J. & Bar-Joseph, Z. Interactive single-cell data analysis using Cellar. Nat Commun 13, (2022).
Moreno, P. et al. User-friendly, scalable tools and workflows for single-cell RNA-seq analysis. Nature Methods vol. 18 327–328 Preprint at https://doi.org/10.1038/s41592-021-01102-w (2021).
Long, Y. et al. Spatially informed clustering, integration, and deconvolution of spatial transcriptomics with GraphST. Nat Commun 14, 1155 (2023).
Li, M. et al. DISCO: A database of Deeply Integrated human Single-Cell Omics data. Nucleic Acids Res 50, D596–D602 (2022).
Korotkevich, G. et al. Fast gene set enrichment analysis. doi:10.1101/060012.
Polański, K. et al. BBKNN: Fast batch alignment of single cell transcriptomes. Bioinformatics 36, 964–965 (2020).
Tran, H. T. N. et al. A benchmark of batch-effect correction methods for single-cell RNA sequencing data. Genome Biol 21, (2020).
Stuart, T. et al. Comprehensive Integration of Single-Cell Data. Cell 177, 1888-1902.e21 (2019).
Argelaguet, R. et al. MOFA+: A statistical framework for comprehensive integration of multi-modal single-cell data. Genome Biol 21, (2020).
Stuart, T., Srivastava, A., Madad, S., Lareau, C. A. & Satija, R. Single-cell chromatin state analysis with Signac. Nat Methods 18, 1333–1341 (2021).
Hafemeister, C. & Satija, R. Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Genome Biol 20, (2019).
Blondel, V. D., Guillaume, J.-L., Lambiotte, R. & Lefebvre, E. Fast unfolding of communities in large networks. (2008) doi:10.1088/1742-5468/2008/10/P10008.
Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat Methods 16, 1289–1296 (2019).
Zhang, F., Wu, Y. & Tian, W. A novel approach to remove the batch effect of single-cell data. Cell Discovery vol. 5 Preprint at https://doi.org/10.1038/s41421-019-0114-x (2019).
Luecken, M. D. et al. Benchmarking atlas-level data integration in single-cell genomics. Nat Methods 19, 41–50 (2022).
Subramanian, A. et al. Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. www.pnas.orgcgidoi10.1073pnas.0506580102 (2005).
Efremova, M., Vento-Tormo, M., Teichmann, S. A. & Vento-Tormo, R. CellPhoneDB: inferring cell–cell communication from combined expression of multi-subunit ligand–receptor complexes. Nat Protoc 15, 1484–1506 (2020).

Tables 1 to 5 are available in the Supplementary Files section.

There is NO Competing Interest.

ezSingleCellSupplfigureS1.pdf
Figure Suppl. S1. Interface of ezSingleCell webserver
ezSingleCellSupplfigureS2.pdf
Figure Suppl. S2. Overview of the 5 Modules in ezSingleCell webserver namely single-cell RNA-seq module, Data Integration module, Spatial Transcriptomics module, single-cell Multiomics module, and single-cell ATAC-seq module
ezSingleCellSupplfigureS3.pdf
Figure Suppl. S3. ezSingleCell scRNA-seq workflow comprising of quality control, comparison of celltypes identified using CELLiD with the ground truth annotation and the marker genes visualization using Violin plot, Feature plot and Ridge plot for certain marker genes for cell types such as MS4A1 (Bcell), KLRB1 (MAIT), CD8A, CD8B (CD8 T cells)
ezSingleCellSupplfigureS4.pdf
Figure Suppl. S4. ezSingleCell data integration workflow comprising of cell type identification and data visualization (violin plot, feature plot and ridge plot) using batch corrected values from the top commonly used methods such as Seurat and Harmony
ezSingleCellSupplfigureS5.pdf
Figure Suppl. S5. ezSingleCell spatial transcriptomics workflow comprising of: (a) Quality control, (b) Visualization of scRNA-seq reference data for deconvolution, (c) Spatial domain visualization, (d) Horizontal integration of spatial datasets (under development)
ezSingleCellTable1.pdf
Table 1. List of publicly available single cell web servers
ezSingleCellTable2.pdf
Table 2. Comparison of ezSingleCell with existing single-cell webservers
ezSingleCellTable3.pdf
Table 3. Comparison of ezSingleCell with popular commercial services
ezSingleCellTable4.pdf
Table 4. List of tools used in ezSingleCell
ezSingleCellTable5.pdf
Table 5. List of datasets used in ezSingleCell
GraphicalAbstract.png

Download PDF

Journal Publication

published 03 Jul, 2024

Read the published version in Nature Communications →

Version 1

posted

You are reading this latest preprint version

ezSingleCell: An integrated one-stop single-cell and spatial omics analysis platform for bench scientists

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Results

Overview of ezSingleCell and its advantages over other tools

ezSingleCell’s scRNA-seq module streamlines the analysis of scRNA-seq data

ezSingleCell’s scIntegration module performs batch correction of multiple scRNA-seq datasets

ezSingleCell’s ST module performs clustering, integration, and deconvolution of spatial transcriptomics

ezSingleCell’s scMultiomics module performs joint analysis of multiple modalities

ezSingleCell’s scATAC-Seq module performs scATAC-seq data analysis and integration

Discussion

Material And Methods

Declarations

References

tables

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1