Systems immunology approaches aim to explore and understand the complexity of the immune system. However, with 350 cluster of differentiation (CD) antigens, over 100 cytokines and chemokines, and many different cell subsets, this is a challenging task (Davis et al., 2017). Liquid mass cytometry (LMC), imaging mass cytometry (IMC), and flow cytometry (FC) are powerful techniques applied to exploratory immunophenotyping and biomarker discovery. With the ability to profile over 40 markers on an individual cell, these techniques have rapidly expanded our understanding of the immune system and its perturbations throughout disease pathogenesis (Bertolo et al., 2020; Wang et al., 2020). However, such comprehensive analyses produce large amounts of data and increased dimensionality, resulting in a demanding computational task to analyze (Newell and Cheng, 2016). Often, a limiting factor to these analyses is the requirement of an in-depth knowledge of computational biology. There is an unmet need for an easy-to-use flexible open-source computational framework to explore a variety of high-dimensional single-cell cytometry datasets such as LMC, IMC, and FC within a single framework for nonspecialists. A number of previous computational cytometry workflows have been proposed, including CyTOF workflow (Nowicka, 2019), (Diggins et al., 2015) and CapX (Ashhurst, 2019). Subsequently, some of these workflows have been incorporated into graphical user interfaces (GUI) or code-based workflow packages such as CATALYST (Crowell et al., 2020), Cytofkit (Chen et al., 2016), diffcyt (Weber et al., 2019), and Spectre (Ashhurst, 2020). However, these approaches can be challenging to implement without significant computational knowledge. We have therefore developed ImmunoCluster, building on and extending the framework proposed in the CyTOF workflow (Nowicka, 2019), to create modular, flexible, and easy-to-use implementations of cytometry analysis pipelines. The ImmunoCluster package is a framework that provides appropriate data structures and methods to increase the utility of these high-dimensional methods and make data analysis and interpretation accessible to all researchers to facilitate immune phenotyping projects, such as those associated with preclinical and clinical studies.

Here, we present ImmunoCluster, an open-source computational framework for the analysis of high-dimensional LMC, IMC, and FC datasets. ImmunoCluster is an R package and framework that focuses on the organization and visualization of data to help define biological identity of cells, construct a hierarchy of biologically meaningful cell populations, and detect significant changes between conditions, timepoints, and groups, including serial samples from a clinical trial setting. The framework uses state-of-the-art computational techniques in an easily amendable and flexible format. Users can manipulate figures and outputs to suit their specific needs. The computational approaches used to analyze high-dimensional data are rapidly evolving, and ImmunoCluster has the flexibility to incorporate these novel computational methods in dimensionality reduction, clustering, and trajectory analysis within its framework. The computational framework is designed for ease of use and broad applicability so that it is simple enough to be implemented by users with only a basic knowledge of R, as well as possessing the flexibility for more advanced users to incorporate additional new methodology and build on the framework. The framework relies on the SingleCellExperiment class (Lun, 2017), which means that the flow cytometry standard (FCS) data is contained within a purpose-built object that stores all stages of analysis to permit multiple analysis paths to be performed in parallel. The framework incorporates methods for applying popular R packages (i.e., RPhenograph [Chen and Chen, 2015], Rtsne [Krijthe, 2018], and FlowSOM [Van Gassen, 2015a]) along with convenient ‘wrapper functions’ allowing for the use these popular dimensionality reduction and clustering algorithms in an easy format for nonspecialists to rapidly produce interpretable data that are extendable to advanced experimental designs. ImmunoCluster also contains functionality for adaptive downsampling at the import and dimensionality reduction stages of the pipeline to overcome significant increases in runtime associated with large datasets in dimensionality reduction algorithms like UMAP and tSNE. The outputs from ImmunoCluster are created using the ggplot package, which generates ggplot graphical objects that are flexible, can be further modified and utilized directly into figures for reports and publications. The core of the visualization tool was developed using scDataviz, a Bioconductor package (Blighe, 2020a) for visualizing single-cell data and influenced by the visual framework outlined in CyTOF workflow (Nowicka, 2019). ImmunoCluster offers an alternative to the currently used frameworks such as CATALYST (Crowell et al., 2020) and Cytofkit (Chen et al., 2016), with a focus on ease of use for biologists, as well as offering additional novel aspects for its users compared with other published frameworks. Importantly, and unique to ImmunoCluster, the scope of functionality in this package also enables the analysis of IMC data. Users are able to run IMC data in line with FC and LMC data analyses. Due to the lower resolution of IMC data (compared with FC and LMC), we specifically designed the ‘ranked expression heatmap’ function to work with IMC data, which allows users to rank markers measured to help identify cell populations. IMC provides a means to analyze the spatial dimension of the cells in situ within tissues, which can provide important insight when ascribing functionality to cell populations. Methods for analyzing large cytometry datasets, particularly IMC datasets, in an open-source computational environment are currently limited. ImmunoCluster has been designed for use by a nonspecialist; however, it would be useful to a range of users from wet-lab nonspecialist experimentalists to experienced computational biologists with potential utility in day-to-day research through to large-scale studies involving multiple longitudinal serial biopsies where samples need to be analyzed and compared.

Clinical immune monitoring is an area of increasing utilization (Hernandez et al., 2020). In the current work, we will present examples of ImmunoCluster’s ability to detect clinically relevant perturbations in immune cellular heterogeneity in LMC, IMC, and FC datasets in human patient and/or healthy samples. ImmunoCluster is easily extensible to several arbitrary experimental designs and can be utilized to aid both discovery of novel subpopulations of cells within heterogeneous samples or as a foundation for monitoring longitudinal changes or responses to treatment. As such, ImmunoCluster provides a resource that will permit unsupervised high-dimensional data analysis to be more widely adopted in immunobiology and many other disciplines.

This section aims to convey the scope of ImmunoCluster, its versatility, and applicability to complex datasets across a range of high-dimensional approaches. We implemented the ImmunoCluster framework for three different types of high-dimensional single-cell data (LMC, IMC, and FC). ImmunoCluster also offers user-friendly flexibility throughout the framework, and researchers can easily adapt all figures and outputs to suit their specific needs, resulting in an abundance of tailored outputs for the user to assess and use in publications, reports, and presentations. An example detailing a variety of tailor-made figures produced by ImmunoCluster can be found in Figure 1—figure supplement 1.

We first used previously published and publicly available LMC data (Hartmann et al., 2019). This dataset was chosen as it is representative of an intricate immunophenotyping project and would test the ability of ImmunoCluster to reproduce published results. Secondly, we investigated two novel IMC datasets, from HNSCC and diffuse large B-cell lymphoma (DLBCL) patients with the aim of demonstrating the applicability of the ImmunoCluster tool for IMC data analysis and cell cluster identification and visualization. Finally, FC data from the BM of seven healthy donors (HDs) during hip surgery were used to show the ability of the framework to analyze conventional FC data and identify rare immune cell populations.

Liquid mass cytometry

The chosen dataset (Hartmann et al., 2019) includes mass cytometry data from 15 patients with leukemia who all received bone marrow transplant (BMT), three of whom developed acute graft versus host disease (GvHD). Samples were taken at 30 and 90 days post-BMT. Thirty-three cell markers, both lineage and functional, were analyzed using mass cytometry (Helios CyTOF system). Prior to uploading into the ImmunoCluster framework, the FCS files were normalized and gated as described in the Materials and methods section. We observed that a typical pipeline run on the full 2.3 million cell LMC GvHD dataset (with UMAP downsampling to 500k cells) would take ~110 min on a 2.9 GHz Intel Core i7 MacBook pro with 16 GB RAM (Figure 1—figure supplement 2).

An initial visualization of the data was first carried out using multidimensional scaling (MDS) plots, which were overlaid with metadata to identify the similarity of patients by condition and day of measurement (Figure 1A, B). A heatmap was also created to give an overview of marker expression for each patient and annotated with the metadata provided (Figure 1C). The UMAP algorithm was used for dimensionality reduction of data. There was a difference in the distribution of cells from GvHD and ‘none’ across the cell islands between both timepoints, 30 and 90 days after BMT (Figure 2A).

Figure 1 with 2 supplements see all

Download asset Open asset

Figure 2 with 3 supplements see all

Download asset Open asset

The mass cytometry panel used in this experiment was specifically designed to identify all major human immune cell lineages (Hartmann et al., 2019); the marker expression across cell islands was visualized and indicated cell types present within the cell islands (Figure 2B) (expression of all markers measured; Figure 2—figure supplement 1). To further distinguish the cell types that were present in the cell islands and visible from the dimensionality reduced UMAP plots (Figure 2B), the FlowSOM clustering algorithm with consensus clustering metaclustering was applied, which has been demonstrated to scale well to large cytometry datasets (Weber and Robinson, 2016). The number of clusters input into the FlowSOM algorithm was slightly higher than the number of expected cell clusters (n = 56) (Figure 2—figure supplement 2). The SingleCellExperiment object stored the data for each K meta-clustering (K1-K56); therefore, further downstream data exploration was carried out looking at different numbers of K clusters. A heatmap showing the expression of the 33 markers measured was used to identify cluster cell types (Figure 2—figure supplement 2), and ImmunoCluster was successfully able to replicate the findings from Hartmann et al., 2019, with 24 of the cell populations being identified (Figure 2C, D). Selecting the correct number of clusters to input into the clustering algorithm can be a challenging aspect of these analyses, but the framework provides the user with multiple parameters to aid this decision. We recommend that users should always overestimate the number of cell populations as data for each K number of clusters will be stored in the SingleCellExperiment object and can therefore be explored and refined later using the different visualization techniques available within the framework. Additionally, the elbow plot generated by the ConsensusClusterPlus package can be examined alongside the clustering algorithm output to guide the decision on the best number of clusters for downstream analysis (Figure 8—figure supplement 6). Over-clustering can allow for the identification of rare cell types within a dataset at the expense of often generating several clusters of highly prevalent cell types that likely represent the same biological cell type. ImmunoCluster provides the tools to manually and reproducibly merge clusters of the same biological cell identity into one group after over-clustering. These clusters can additionally be merged into each other to use a less granular, higher-level population annotation (see Figure 2—figure supplement 3 for an example of higher-level clustering).

The abundance of all cell types across all samples was measured, and CD14⁺CD16^- monocytes were identified as the most abundant population, correlating with the Hartmann et al. data (Hartmann et al., 2019; Figure 2E). Individually displaying patient’s cell-type abundance means that we could identify variation within a group of interest; for example, we observed that although there is variation within the GvHD and none, overall, they appear to follow the same trends in cell-type abundance (Figure 3A). Additionally, significant differences between memory B-cells (false discovery rate [FDR] p=4.38×10⁻³), naïve B-cells (FDR p=1.35×10⁻²), and naïve CD4⁺ T-cells (FDR p=3.47×10⁻²) were identified between the GvHD and none (Figure 3B, C) using a t-test. A difference in number of naïve B-cells was previously reported by Hartmann et al., 2019 in this dataset and can be seen in Figure 2A (using Figure 2D for cell island identification), where there is a noticeable reduction of these cells in the GvHD patients. In addition, a reduction in naïve CD4⁺ T-cells can also be seen, which was also previously reported, but to a lesser extent. A volcano plot can be used to highlight the differentially abundant cell clusters between GvHD and none (GvHD logFC+ve and none logFC-ve), where cell types with FDR p<0.05 are shown in red (Figure 3D), which is also in line with the previously published data (Hartmann et al., 2019). We compared ImmunoCluster’s differential abundance testing output (stat_test_clust), run in the t-test mode, with the Hartmann et al. data (Hartmann et al., 2019) into the diffcyt computational framework (Weber et al., 2019), which is a state-of-the-art tool for differential discovery analyses. Diffcyt identified the same three cell clusters (memory B-cells, naïve B-cells, and naïve CD4⁺ T-cells) as differentially abundant (FDR p<0.05) between the GvHD and none conditions. This demonstrates concordance with ImmunoCluster’s statistical output that identified naïve B-cells and naïve CD4⁺ T-cells as one of the principal differentially abundant populations, which was also identified in the original analysis of the data (Supplementary file 4). Additionally, we explored the expression of checkpoint-related molecules, and their receptors, markers of proliferation, and invariant natural killer T (iNK T) cells in CD8⁺ T-cells and compared the expression of these in the patients with GvHD and none (Figure 3E). These markers help assess the functional states of cells and the checkpoint-related molecules and proliferative activity markers such as PD-1, PD-L1, TIM3, Ki-67, and TCRVa24-Ja18; the expression of these antigens has been proposed previously as candidate biomarkers for immunotherapy (Hartmann et al., 2019). Visually, GvHD patients had a higher expression of PD-1 and a noticeable difference in the Ki-67 proliferation marker as expected for these patients (Figure 3E).

Figure 3 with 1 supplement see all

Download asset Open asset

Head and neck squamous cell carcinoma tissue section

Cell segmented regions are annotated in Figure 4A, where regions 1–3 were selected as tumor regions and 3–6 as stroma regions. The ImmunoCluster tool was able to distinguish between the stroma and tumor regions of the tissue section from the HNSCC tumor (Figure 4B), and with additional markers could provide a more in-depth analysis of the cell phenotypes in each respective region of the tumor. Dimensionality reduction of the data clearly separated the cell islands belonging to the stroma and tumor regions of the tissue (Figure 4C), suggesting different cell phenotypes between the regions. The FlowSOM algorithm was applied to cluster the cells and identify cell populations. The rank heatmap was used to show the ranked expression of each marker (ranked 1–10, with 10 being high) for each cluster (1–10) (Figure 4—figure supplement 1), and the identified cell types were then used to annotate the heatmap (Figure 4D). High expression of E-cadherin can be used as a marker of cancerous tissue in HNSCC, and its expression correlates with the cancer regions selected for analysis (Figure 4D). We focused on CD8⁺ T-cell, B-cell, and macrophage cell clusters (Hladíková et al., 2019; Evrard et al., 2019). Regions 2 and 3 of the tumor mostly consist of proliferating (Ki-67⁺) tumor cells and macrophages (tumor-associated and PD-L1⁺). Region 1 was a mix of immune cells, B-cells, and PD-L1⁺ CD8⁺ T-cells (Figure 4E). Stromal regions 5 and 6 are mostly PD-L1⁺ macrophages and CD8⁺ T-cells. Region 4 has a cluster of mixed immune cells, and from the IMC image (Figure 4A) we can see that this region looks like the tumor is encroaching into the stromal region. The stroma seems to be mostly nonproliferating cells (Ki-67^- compared to the tumor region) and CD8⁺ T-cells, including a CD20⁺ CD8⁺ cytotoxic subset (Gingele et al., 2020).

Figure 4 with 1 supplement see all

Download asset Open asset

Diffuse large B-cell lymphoma lymph node section

The IMC DLBCL lymph node section was split by high and low Ki-67 expression to identify the highly proliferative tumor cells (Figure 5A); Ki-67 is used as a prognostic marker in DLBCL (Tang et al., 2017). The FlowSOM algorithm was applied to the data to split the IMC data into clusters; these clusters were projected onto the UMAP plot to visualize how these clusters were split between the Ki-67 high and low (Figure 5B). The marker expression ranking tool was applied to the clustered data and used to identify cell types (Figure 5C, Figure 5—figure supplement 1). The majority of Ki-67 high population were proliferating tumor cells (84%), and the Ki-67 low population consisted of a heterogeneous collection of immune cell populations, such as CD4⁺ T-cells, CD8⁺ T-cells, B-cells, macrophages, and dendritic cells (DCs) that were successfully identified using the ImmunoCluster tool (Figure 5D).

Figure 5 with 1 supplement see all

Download asset Open asset

Flow cytometry

The HD BM taken during hip surgery was gated for the CD3⁺CD4⁺ population before analysis within the ImmunoCluster framework, with the aim of testing ImmunoCluster’s ability to identify minor populations such as Tregs and their subpopulations (Kordasti et al., 2016). The FlowSOM algorithm was applied to the data, resulting in 40 clusters (Figure 6—figure supplement 1), and a heatmap was created with median expression of markers to identify cell types (Figure 6B). Due to the markers used and the low prevalence of Tregs, the majority of cells were identified as CD4⁺ T-cells. Additionally, three populations of Tregs were identified by the ImmunoCluster tool: Tregs (CD25⁺CD127^low) (3.6%, 1.3–5.5), Treg A (CD25⁺CD127^low CD45RA⁺) (0.7%, 0.1–2.0), and Treg B-cells (CD25⁺CD127^lowCCR4⁺CD95^high) (0.9%, 0.2–2.4) (Figure 6C). The abundance of Treg A and B for all HDs is shown in Figure 6D, and their distribution was as expected (Kordasti et al., 2016).

Figure 6 with 1 supplement see all

Download asset Open asset

Here, we provide an overview of the flexibility and scope for utilizing ImmunoCluster, an open-access easy-to-use framework for LMC, IMC, and FC dataset analyses. The ImmunoCluster package emphasizes ease of use in the generation of flexible and modular cytometry analysis pipelines for applications such as clinical immune monitoring studies. Explanations of the functions and workflow have been provided, and an in-depth step-by-step protocol and tutorial are available on the GitHub site (https://github.com/kordastilab/ImmunoCluster).

The purpose for creating the ImmunoCluster framework was to support nonspecialist researchers to carry out complex immunophenotyping experiments. The dawn of ‘cytometry big data’ and its ever-growing utilization has surpassed the number of experimental parameters that a researcher can feasibly analyze as a collective. The ImmunoCluster framework was designed in collaboration between wet lab and computational biologists, with the purpose of creating a tool that enables state-of-the-art analysis, yet easy to use in its application. We provide sufficient detail and explanation for the nonspecialist researcher to understand the need for each analysis step as well as how to confidently execute the process, with an aim of performing transparent and reproducible analyses.

ImmunoCluster has been built with an emphasis on generalizability and scalability to facilitate a broad use and is an advance over other open-source tools as it provides an integrated framework to uniquely perform a complete computational pipeline on high-dimensional LMC, IMC, and FC data. The inclusion of IMC data analysis brings a unique element to the ImmunoCluster framework and differentiates it from other published protocols. ImmunoCluster was built to easily leverage many popular R packages for cytometry data (i.e., RPhenograph [Levine et al., 2015], Rtsne [van der Maaten and Hinton, 2008; Laurens van der, 2014], and FlowSOM [Van Gassen, 2015b]), selected because they are open-source and regularly maintained with extensive documentation. All methods prioritize customizable and attractive visualizations designed to be used by both dry- and wet-lab researchers/clinicians. The inclusion of adaptive downsampling of cells for the running of computationally intensive dimensionality reduction steps, using tools such as UMAP and tSNE, allows ImmunoCluster to generate cytometry analysis pipelines that are readily scalable to a dataset of millions of cells across several samples and a variety of experimental or phenotypic conditions. ImmunoCluster can also be run on a local desktop/laptop computer with standard configuration, depending on the number of cells, adaptive downsampling settings and selection of clustering algorithms can usually be run within a day end-to-end (Figure 1—figure supplement 2). Whilst the PhenoGraph clustering method does not readily scale, the FlowSOM-based clustering method can be implemented on much larger datasets.

The limitations of this computational framework are important to take into consideration when designing an experiment. Unsupervised clustering (stage 2) of samples is the most important stage of the framework, and its ability to accurately define populations across all samples is critical to all later stages of investigation. The number of k clusters to generate (cluster resolution) can significantly affect the ability to identify or determine a change in a biologically relevant population. Merging two subpopulations due to low resolution of clustering may mask an important experimental observation. Determining the precise number of clusters that are relevant for a given dataset is an important step. In our step-by-step guide, we provide complete documentation and clear user-friendly approaches to optimally define this.

Technical inter-sample marker signal variability due to batch effects may impact the ability of unsupervised analysis to reliably detect certain populations if significant batch effects are present. MDS plots created in stage 1 of the framework can be used to detect batch effects as well as other technical artifacts such as antibody staining anomalies. An approach commonly employed in mass cytometry to overcome this is to use sample barcoding, reducing the variability between each sample, allowing all samples to be exposed to the same antibody mixture (Schulz and Mei, 2019). Another regularly employed approach is the inclusion of a shared control sample in each independent batch, for example, the same cell type as the experimental samples but all from the same HD. Statistical methods for batch correction may also be applied; in recent years, a class of methods called remove unwanted variation (RUV) have been developed, and CytofRUV (Trussart et al., 2020) is a recently developed package specifically designed for CyTOF dataset batch correction. If there are significant differences in the total number of cells recovered between samples, samples with many more cells may bias the clustering. Samples with very few cells recovered may result in information loss and missing populations that are in fact present. The ImmunoCluster framework provides users with two opportunities (creating the SingeCellExperiment and running the dimensionality reduction) to downsample, which means the same number of samples will be used for each sample, rectifying the problem of varying cell numbers. If a particular sample has very few cells, they may need to be excluded from the analysis as they may not be a representative sample. Although channel spillover is diminished in mass cytometry, it still exists in fluorescence-based FC and should be considered when designing antibody panels to reduce the effects on introducing cell phenotype artifacts in downstream unsupervised analysis. As such, initial exploratory data analysis (e.g., MDS plots) is key to determining if any of these confounding factors might be present in the data. Additionally, users may want to introduce fluorescence‐minus‐one controls, where cells are stained with all fluorescently tagged antibodies except for the one of interest (Roederer, 2001).

Staining protocol for a human tissue section taken from a patient with head and neck squamous cell carcinoma (HNSCC).

The liquid mass cytometry dataset is available from FlowRepository (http://flowrepository.org/id/FR-FCM-Z244), the additional three datasets: 'Imaging mass cytometry data: Head and neck squamous cellcarcinoma tissue section', 'Flow cytometry data: healthy donor bone marrow taken during hip surgery', and 'Imaging mass cytometry data: Head and neck squamous cell carcinoma tissue section', are available from Dryad (https://datadryad.org/stash).

1. 1. Opzoomer JW
   2. Timms J
   3. Blighe K
   4. Mourikis TP
   5. Chapuis N
   6. Bekoe R
   7. Kareemaghay S
   8. Nocerino P
   9. Apollonio B
   10. Ramsay AG
   11. Tavassoli M
   12. Harrison C
   13. Ciccarelli F
   14. Parker P
   15. Fontenay M
   16. Barber PR
   17. Arnold JN
   18. Kordasti S

   (2021) Dryad

   Imagingmass cytometry data: Head and neck squamous cell carcinoma tissuesection.

   https://doi.org/10.5061/dryad.gf1vhhmpr
2. 1. Opzoomer JW
   2. Timms J
   3. Blighe K
   4. Mourikis TP
   5. Chapuis N
   6. Bekoe R
   7. Kareemaghay S
   8. Nocerino P
   9. Apollonio B
   10. Ramsay AG
   11. Tavassoli M
   12. Harrison C
   13. Ciccarelli F
   14. Parker P
   15. Fontenay M
   16. Barber PR
   17. Arnold JN
   18. Kordasti S

   (2021) Dryad

   Flow cytometry data: healthy donor bone marrow taken during hipsurgery.

   https://doi.org/10.5061/dryad.4b8gthtcf
3. 1. Opzoomer JW
   2. Timms J
   3. Blighe K
   4. Mourikis TP
   5. Chapuis N
   6. Bekoe R
   7. Kareemaghay S
   8. Nocerino P
   9. Apollonio B
   10. Ramsay AG
   11. Tavassoli M
   12. Harrison C
   13. Ciccarelli F
   14. Parker P
   15. Fontenay M
   16. Barber PR
   17. Arnold JN
   18. Kordasti S

   (2021) Dryad

   Imaging mass cytometry data: Diffuse large B-cell lymphoma lymph node section.

   https://doi.org/10.5061/dryad.3n5tb2rhd

1. 1. Hartmann FJ
   2. Babdor J
   3. Gherardini PF
   4. Amir ED
   5. Jones K
   6. Sahaf B
   7. Marquez DM
   8. Krutzik P
   9. O'Donnell E
   10. Sigal N
   11. Maecker HT
   12. Meyer E
   13. Spitzer MH
   14. Bendall SC

   (2019) Flow Repository

   ID FR-FCM-Z244. Identification of disease-associated immune signatures following bone marrow transplantation.

   http://flowrepository.org/id/FR-FCM-Z244

1. 1. Angerer P
   2. Haghverdi L
   3. Büttner M
   4. Theis FJ
   5. Marr C
   6. Buettner F

   (2016) Destiny: diffusion maps for large-scale single-cell data in R

   Bioinformatics 32:1241–1243.

   https://doi.org/10.1093/bioinformatics/btv715

   • PubMed
   • Google Scholar
2. Book

   1. Ashhurst TM

   (2019) Analysis of the Murine Bone Marrow Hematopoietic System Using Mass and Flow Cytometry

   In: McGuire H. M, Ashhurst T. M, editors. Mass Cytometry: Methods and Protocols. Springer. pp. 159–192.

   https://doi.org/10.1007/978-1-4939-9454-0

   • Google Scholar
3. 1. Ashhurst TM

   (2020) Integration, exploration, and analysis of high-dimensional single-cell cytometry data using spectre

   Cytometry. Part a : The Journal of the International Society for Analytical Cytology 10:e24350.

   https://doi.org/10.1002/cyto.a.24350

   • Google Scholar
4. 1. Becht E
   2. McInnes L
   3. Healy J
   4. Dutertre C-A
   5. Kwok IWH
   6. Ng LG
   7. Ginhoux F
   8. Newell EW

   (2019) Dimensionality reduction for visualizing single-cell data using UMAP

   Nature Biotechnology 37:38–44.

   https://doi.org/10.1038/nbt.4314

   • Google Scholar
5. 1. Bertolo M
   2. Baumgart S
   3. Durek P
   4. Peddinghaus A
   5. Mei H
   6. Rose T
   7. Enghard P
   8. Grützkau A

   (2020) Deep phenotyping of urinary leukocytes by mass cytometry reveals a leukocyte signature for early and Non-Invasive prediction of response to treatment in active lupus nephritis

   Frontiers in Immunology 11:256.

   https://doi.org/10.3389/fimmu.2020.00256

   • PubMed
   • Google Scholar
6. Software

   1. Blighe K

   (2020a) scDataviz: Single Cell Dataviz and Downstream Analyses

   scDataviz: Single Cell Dataviz and Downstream Analyses.

   https://github.com/kevinblighe/scDataviz
7. Software

   1. Blighe AL

   (2020b) PCAtools: PCAtools: Everything Principal Components Analysis

   PCAtools: PCAtools: Everything Principal Components Analysis.

   https://github.com/kevinblighe/PCAtools
8. 1. Chen H
   2. Lau MC
   3. Wong MT
   4. Newell EW
   5. Poidinger M
   6. Chen J

   (2016) Cytofkit: a bioconductor package for an integrated mass cytometry data analysis pipeline

   PLOS Computational Biology 12:e1005112.

   https://doi.org/10.1371/journal.pcbi.1005112

   • PubMed
   • Google Scholar
9. Software

   1. Chen J
   2. Chen H

   (2015) Rphenograph

   Rphenograph.

   https://github.com/JinmiaoChenLab/Rphenograph
10. 1. Chen TJ
    2. Kotecha N

    (2014) Cytobank: providing an analytics platform for community cytometry data analysis and collaboration

    Current Topics in Microbiology and Immunology 377:127–157.

    https://doi.org/10.1007/82_2014_364

    • PubMed
    • Google Scholar
11. Software

    1. Crowell H
    2. Chevrier S
    3. Robinson M

    (2020) CATALYST: Cytometry dATa anALYSis Tools

    CATALYST: Cytometry dATa anALYSis Tools.

    https://github.com/HelenaLC/CATALYST
12. 1. Davis MM
    2. Tato CM
    3. Furman D

    (2017) Systems immunology: just getting started

    Nature Immunology 18:725–732.

    https://doi.org/10.1038/ni.3768

    • PubMed
    • Google Scholar
13. 1. Diggins KE
    2. Ferrell PB
    3. Irish JM

    (2015) Methods for discovery and characterization of cell subsets in high dimensional mass cytometry data

    Methods 82:55–63.

    https://doi.org/10.1016/j.ymeth.2015.05.008

    • PubMed
    • Google Scholar
14. 1. DiGiuseppe JA
    2. Cardinali JL
    3. Rezuke WN
    4. Pe'er D

    (2018) PhenoGraph and viSNE facilitate the identification of abnormal T-cell populations in routine clinical flow cytometric data

    Cytometry Part B: Clinical Cytometry 94:744–757.

    https://doi.org/10.1002/cyto.b.21588

    • Google Scholar
15. 1. Evrard D
    2. Szturz P
    3. Tijeras-Raballand A
    4. Astorgues-Xerri L
    5. Abitbol C
    6. Paradis V
    7. Raymond E
    8. Albert S
    9. Barry B
    10. Faivre S

    (2019) Macrophages in the microenvironment of head and neck Cancer: potential targets for Cancer therapy

    Oral Oncology 88:29–38.

    https://doi.org/10.1016/j.oraloncology.2018.10.040

    • PubMed
    • Google Scholar
16. 1. Finak G
    2. Frelinger J
    3. Jiang W
    4. Newell EW
    5. Ramey J
    6. Davis MM
    7. Kalams SA
    8. De Rosa SC
    9. Gottardo R

    (2014) OpenCyto: an open source infrastructure for scalable, robust, reproducible, and automated, end-to-end flow cytometry data analysis

    PLOS Computational Biology 10:e1003806.

    https://doi.org/10.1371/journal.pcbi.1003806

    • PubMed
    • Google Scholar
17. 1. Gingele S
    2. Skripuletz T
    3. Jacobs R

    (2020) Role of CD20(+) T cells in multiple sclerosis: implications for treatment with ocrelizumab

    Neural Regeneration Research 15:663–664.

    https://doi.org/10.4103/1673-5374.266913

    • PubMed
    • Google Scholar
18. Software

    1. Hammill D

    (2020)

    CytoExploreR: Interactive Analysis of Cytometry Data

    CytoExploreR: Interactive Analysis of Cytometry Data.
19. 1. Hartmann FJ
    2. Babdor J
    3. Gherardini PF
    4. Amir ED
    5. Jones K
    6. Sahaf B
    7. Marquez DM
    8. Krutzik P
    9. O'Donnell E
    10. Sigal N
    11. Maecker HT
    12. Meyer E
    13. Spitzer MH
    14. Bendall SC

    (2019) Comprehensive immune monitoring of clinical trials to advance human immunotherapy

    Cell Reports 28:819–831.

    https://doi.org/10.1016/j.celrep.2019.06.049

    • PubMed
    • Google Scholar
20. 1. Hernandez C
    2. Arasanz H
    3. Chocarro L
    4. Bocanegra A
    5. Zuazo M
    6. Fernandez-Hinojal G
    7. Blanco E
    8. Vera R
    9. Escors D
    10. Kochan G

    (2020) Systemic blood immune cell populations as biomarkers for the outcome of immune checkpoint inhibitor therapies

    International Journal of Molecular Sciences 21:2411.

    https://doi.org/10.3390/ijms21072411

    • Google Scholar
21. 1. Hladíková K
    2. Koucký V
    3. Bouček J
    4. Laco J
    5. Grega M
    6. Hodek M
    7. Zábrodský M
    8. Vošmik M
    9. Rozkošová K
    10. Vošmiková H
    11. Čelakovský P
    12. Chrobok V
    13. Ryška A
    14. Špíšek R
    15. Fialová A

    (2019) Tumor-infiltrating B cells affect the progression of oropharyngeal squamous cell carcinoma via cell-to-cell interactions with CD8⁺ T cells

    Journal for ImmunoTherapy of Cancer 7:261.

    https://doi.org/10.1186/s40425-019-0726-6

    • PubMed
    • Google Scholar
22. 1. Kordasti S
    2. Costantini B
    3. Seidl T
    4. Perez Abellan P
    5. Martinez Llordella M
    6. McLornan D
    7. Diggins KE
    8. Kulasekararaj A
    9. Benfatto C
    10. Feng X
    11. Smith A
    12. Mian SA
    13. Melchiotti R
    14. de Rinaldis E
    15. Ellis R
    16. Petrov N
    17. Povoleri GA
    18. Chung SS
    19. Thomas NS
    20. Farzaneh F
    21. Irish JM
    22. Heck S
    23. Young NS
    24. Marsh JC
    25. Mufti GJ

    (2016) Deep phenotyping of tregs identifies an immune signature for idiopathic aplastic Anemia and predicts response to treatment

    Blood 128:1193–1205.

    https://doi.org/10.1182/blood-2016-03-703702

    • PubMed
    • Google Scholar
23. Software

    1. Krijthe J

    (2018) Rtsne

    Rtsne.

    https://github.com/jkrijthe/Rtsne
24. 1. Laurens van der M

    (2014)

    Accelerating t-SNE using Tree-Based algorithms

    Journal of Machine Learning Research 15:3221–3245.

    • Google Scholar
25. 1. Levine JH
    2. Simonds EF
    3. Bendall SC
    4. Davis KL
    5. Amir AD
    6. Tadmor MD
    7. Litvin O
    8. Fienberg HG
    9. Jager A
    10. Zunder ER
    11. Finck R
    12. Gedman AL
    13. Radtke I
    14. Downing JR
    15. Pe'er D
    16. Nolan GP

    (2015) Data-Driven phenotypic dissection of AML reveals Progenitor-like cells that correlate with prognosis

    Cell 162:184–197.

    https://doi.org/10.1016/j.cell.2015.05.047

    • PubMed
    • Google Scholar
26. Software

    1. Lun A

    (2017)

    SingleCellExperiment: S4 Classes for Single Cell Data

    SingleCellExperiment: S4 Classes for Single Cell Data.
27. 1. Lun ATL
    2. Richard AC
    3. Marioni JC

    (2017) Testing for differential abundance in mass cytometry data

    Nature Methods 14:707–709.

    https://doi.org/10.1038/nmeth.4295

    • PubMed
    • Google Scholar
28. 1. Newell EW
    2. Cheng Y

    (2016) Mass cytometry: blessed with the curse of dimensionality

    Nature Immunology 17:890–895.

    https://doi.org/10.1038/ni.3485

    • PubMed
    • Google Scholar
29. Software

    1. Nowicka M

    (2019)

    CyTOF Workflow: Differential Discovery in High-Throughput High-Dimensional Cytometry Datasets

    CyTOF Workflow: Differential Discovery in High-Throughput High-Dimensional Cytometry Datasets.
30. 1. Roederer M

    (2001) Spectral compensation for flow cytometry: visualization artifacts, limitations, and caveats

    Cytometry 45:194–205.

    https://doi.org/10.1002/1097-0320(20011101)45:3<194::AID-CYTO1163>3.0.CO;2-C

    • PubMed
    • Google Scholar
31. 1. Schulz D
    2. Zanotelli VRT
    3. Fischer JR
    4. Schapiro D
    5. Engler S
    6. Lun XK
    7. Jackson HW
    8. Bodenmiller B

    (2018) Simultaneous multiplexed imaging of mRNA and proteins with subcellular resolution in breast Cancer tissue samples by mass cytometry

    Cell Systems 6:25–36.

    https://doi.org/10.1016/j.cels.2017.12.001

    • PubMed
    • Google Scholar
32. 1. Schulz AR
    2. Mei HE

    (2019) Surface barcoding of live PBMC for multiplexed mass cytometry

    Methods in Molecular Biology 1989:93–108.

    https://doi.org/10.1007/978-1-4939-9454-0_7

    • PubMed
    • Google Scholar
33. 1. Spidlen J
    2. Breuer K
    3. Rosenberg C
    4. Kotecha N
    5. Brinkman RR

    (2012) FlowRepository: a resource of annotated flow cytometry datasets associated with peer-reviewed publications

    Cytometry Part A 81A:727–731.

    https://doi.org/10.1002/cyto.a.22106

    • Google Scholar
34. 1. Tang YL
    2. Zhou Y
    3. Cheng LL
    4. Su YZ
    5. Wang CB

    (2017) BCL2/Ki-67 index predict survival in germinal center B-cell-like diffuse large B-cell lymphoma

    Oncology Letters 14:3767–3773.

    https://doi.org/10.3892/ol.2017.6577

    • PubMed
    • Google Scholar
35. 1. Trussart M
    2. Teh CE
    3. Tan T
    4. Leong L
    5. Gray DH
    6. Speed TP

    (2020) Removing unwanted variation with CytofRUV to integrate multiple CyTOF datasets

    eLife 9:e59630.

    https://doi.org/10.7554/eLife.59630

    • PubMed
    • Google Scholar
36. 1. van der Maaten L
    2. Hinton GJ

    (2008)

    Visualizing data using t-SNE

    Journal of Machine Learning Research 9:2579–2605.

    • Google Scholar
37. Software

    1. Van Gassen S

    (2015a) FlowSOM. Bioconductor Version: Release (3.12)

    FlowSOM. Bioconductor Version: Release (3.12).

    https://github.com/saeyslab/FlowSOM
38. 1. Van Gassen S

    (2015b) FlowSOM: using self-organizing maps for visualization and interpretation of cytometry data

    Cytometry Part A 87:636–645.

    https://doi.org/10.1002/cyto.a.22625

    • Google Scholar
39. 1. Wang W
    2. Khoury JD
    3. Miranda RN
    4. Jorgensen JL
    5. Xu J
    6. Loghavi S
    7. Li S
    8. Pemmaraju N
    9. Nguyen T
    10. Medeiros LJ
    11. Wang SA

    (2020) Immunophenotypic characterization of reactive and neoplastic plasmacytoid dendritic cells permits establishment of a 10-color flow cytometric panel for initial workup and residual disease evaluation of blastic plasmacytoid dendritic cell neoplasm

    Haematologica 106:1047–1055.

    https://doi.org/10.3324/haematol.2020.247569

    • Google Scholar
40. 1. Weber LM
    2. Nowicka M
    3. Soneson C
    4. Robinson MD

    (2019) Diffcyt: differential discovery in high-dimensional cytometry via high-resolution clustering

    Communications Biology 2:183.

    https://doi.org/10.1038/s42003-019-0415-5

    • PubMed
    • Google Scholar
41. 1. Weber LM
    2. Robinson MD

    (2016) Comparison of clustering methods for high-dimensional single-cell flow and mass cytometry data

    Cytometry Part A 89:1084–1096.

    https://doi.org/10.1002/cyto.a.23030

    • Google Scholar
42. Software

    1. Wilkerson M
    2. Waltman P

    (2013)

    ConsensusClusterPlus: ConsensusClusterPlus

    ConsensusClusterPlus: ConsensusClusterPlus.

Author details

1. James W Opzoomer

   School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom

   Contribution

   Conceptualization, Resources, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing

   Contributed equally with

   Jessica A Timms and Kevin Blighe

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-6842-756X

2. Jessica A Timms

   School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom

   Contribution

   Conceptualization, Resources, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing

   Contributed equally with

   James W Opzoomer and Kevin Blighe

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-3687-9312

3. Kevin Blighe

   School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom

   Contribution

   Conceptualization, Resources, Data curation, Software, Formal analysis, Validation, Investigation, Methodology, Writing - original draft, Writing - review and editing

   Contributed equally with

   James W Opzoomer and Jessica A Timms

   Competing interests

   No competing interests declared

4. Thanos P Mourikis

   School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom

   Contribution

   Software, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

5. Nicolas Chapuis

   Institut Cochin, Institut National de la Santé et de la Recherche Médicale U1016, Centre National de la Recherche Scientifique, Unité Mixte de Recherche 8104, Université Paris Descartes, Paris, France

   Contribution

   Data curation, Writing - review and editing

   Competing interests

   No competing interests declared

6. Richard Bekoe

   UCL Cancer Institute, Paul O'Gorman Building, University College London, London, United Kingdom

   Contribution

   Formal analysis

   Competing interests

   No competing interests declared

7. Sedigeh Kareemaghay

   Centre for Host Microbiome Interaction, FoDOCS, King’s College, Guy’s Hospital, London, United Kingdom

   Contribution

   Investigation

   Competing interests

   No competing interests declared

8. Paola Nocerino

   School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom

   Contribution

   Investigation

   Competing interests

   No competing interests declared

9. Benedetta Apollonio

   School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom

   Contribution

   Investigation

   Competing interests

   No competing interests declared

10. Alan G Ramsay

    School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom

    Contribution

    Conceptualization, Resources, Supervision, Funding acquisition, Investigation, Methodology, Writing - original draft, Project administration, Writing - review and editing

    Competing interests

    No competing interests declared

    "This ORCID iD identifies the author of this article:" 0000-0002-0452-0420

11. Mahvash Tavassoli

    Centre for Host Microbiome Interaction, FoDOCS, King’s College, Guy’s Hospital, London, United Kingdom

    Contribution

    Investigation

    Competing interests

    No competing interests declared

12. Claire Harrison

    1. School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom
    2. Haematology Department, Guy’s Hospital, London, United Kingdom

    Contribution

    Funding acquisition, Investigation, Writing - review and editing

    Competing interests

    No competing interests declared

13. Francesca Ciccarelli

    1. School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom
    2. Cancer Systems Biology Laboratory, The Francis Crick Institute, London, United Kingdom

    Contribution

    Funding acquisition, Writing - review and editing

    Competing interests

    No competing interests declared

14. Peter Parker

    1. School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom
    2. Francis Crick Institute, London, United Kingdom

    Contribution

    Writing - review and editing

    Competing interests

    No competing interests declared

15. Michaela Fontenay

    Institut Cochin, Institut National de la Santé et de la Recherche Médicale U1016, Centre National de la Recherche Scientifique, Unité Mixte de Recherche 8104, Université Paris Descartes, Paris, France

    Contribution

    Investigation, Writing - review and editing

    Competing interests

    No competing interests declared

    "This ORCID iD identifies the author of this article:" 0000-0002-5492-6349

16. Paul R Barber

    1. School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom
    2. UCL Cancer Institute, Paul O'Gorman Building, University College London, London, United Kingdom

    Contribution

    Resources, Software, Investigation, Writing - review and editing

    Competing interests

    No competing interests declared

    "This ORCID iD identifies the author of this article:" 0000-0002-8595-1141

17. James N Arnold

    School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom

    Contribution

    Conceptualization, Resources, Software, Supervision, Funding acquisition, Investigation, Methodology, Writing - original draft, Project administration, Writing - review and editing

    Contributed equally with

    Shahram Kordasti

    For correspondence

    james.n.arnold@kcl.ac.uk

    Competing interests

    No competing interests declared

18. Shahram Kordasti

    1. School of Cancer and Pharmaceutical Sciences, King’s College London, Faculty of Life Sciences and Medicine, Guy’s Hospital, London, United Kingdom
    2. Haematology Department, Guy’s Hospital, London, United Kingdom

    Contribution

    Conceptualization, Resources, Supervision, Funding acquisition, Methodology, Writing - original draft, Project administration, Writing - review and editing

    Contributed equally with

    James N Arnold

    For correspondence

    shahram.kordasti@kcl.ac.uk

    Competing interests

    Honoraria: Beckman Coulter, GWT-TUD, Alexion. Consulting or Advisory Role: Syneos Health. Research Funding: Celgene, Novartis pharmaceutical

    "This ORCID iD identifies the author of this article:" 0000-0002-0347-4207

• 3,990

  views

• 417

  downloads

• 12

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.