This tutorial presents an outline of the basic series of steps for analyzing a 10x Genomics Gene Expression with Feature Barcoding (antibody) data set in Partek Flow starting with the output of Cell Ranger.  

If you are starting with the raw data (FASTQ files), please begin with our Processing CITE-Seq data tutorial, which will take you from raw data to count matrix files. 

If you have Cell Hashing data, please see our documentation on Hashtag demultiplexing. 

This tutorial includes only one sample, but the same steps will be followed when analyzing multiple samples. For notes on a few aspects specific to a multi-sample analysis, please see our Single Cell RNA-Seq Analysis (Multiple Samples) tutorial. 

If you are new to Partek Flow, please see Getting Started with Your Partek Flow Hosted Trial for information about data transfer and import and Creating and Analyzing a Project for information about the Partek Flow user interface.  

Data set

The data set for this tutorial is a demonstration data set from 10x Genomics. The sample includes cells from a dissociated Extranodal Marginal Zone B-Cell Tumor (MALT: Mucosa-Associated Lymphoid Tissue) stained with BioLegend TotalSeq-B antibodies. We are starting with the Feature / cell matrix HDF5 (filtered) produced by Cell Ranger. 

Importing feature barcoding data

A Single cell counts data node will be created after the file has been imported.

Split matrix

The Single cell counts data node contains two different types of data, mRNA measurements and protein measurements. So that we can process these two different types of data separately, we will split the data by data type. 

A rectangle, or task node, will be created for Split matrix along with two output circles, or data nodes, one for each data type (Figure 2). The labels for these data types are determined by features.csv file used when processing the data with Cell Ranger. Here, our data is labeled Gene Expression, for the mRNA data, and Antibody Capture, for the protein data. 

 

Filter low-quality cells

An important step in analyzing single cell RNA-Seq data is to filter out low-quality cells. A few examples of low-quality cells are doublets, cells damaged during cell isolation, or cells with too few reads to be analyzed. In a CITE-Seq experiment, protein aggregation in the antibody staining reagents can cause a cell to have a very high number of counts; these are low-quality cells are can be excluded. Additionally, if all cells in a data set are expected to show a baseline level of expression for one of the antibodies used, it may be appropriate to filter out cells with very low counts. You can do this in Partek Flow using the Single cell QA/QC task. 

We will start with the protein data.

This produces a Single-cell QA/QC task node (Figure 4). 

 

The task report lists the number of counts per cell and the number of detected features per cell in two violin plots. For more information, please see our documentation for the Single cell QA/QC task. For this analysis, we will set a maximum counts threshold to exclude potential protein aggregates and, because we expect every cell to be bound by several antibodies, we will also set a minimum counts threshold. 

 

 

The output is a Filtered single cell counts data node (Figure 6).

 

Next, we can repeat this process for the Gene Expression data node. 

This produces a Single-cell QA/QC task node (Figure 7). 

 

The task report lists the number of counts per cell, the number of detected features per cell, and the percentage of mitochondrial reads per cell in three violin plots. For this analysis, we will set a maximum counts threshold maximum and minimum thresholds for total counts and detected genes to exclude potential doublets and a maximum mitochondrial reads percentage filter to exclude potential dead or dying cells. 

 

The output is a Filtered single cell counts data node (Figure 9).

 

Normalization

After excluding low-quality cells, we can normalize the data. 

We will start with the protein data. We will normalize this data using Centered log-ratio (CLR). CLR was used to normalize antibody capture protein counts data in the paper that introudced CITE-Seq (Stoeckius et al. 2017) and in subsequent publications on similar assays (Stoeckiius et al. 2018, Mimitou et al. 2018). CLR normalization includes the following steps: Add 1, Divide by Geometric mean, Add 1, log base e.

Normalization produces a Normalized counts data node on the Antibody Capture branch of the pipeline. 

Next, we can normalize the mRNA data. We will use the recommended normalization method in Partek Flow, which accounts for differences in library size, or the total number of UMI counts, per cell and log transforms the data. To match the CLR normalization used on the Antibody Capture data, we will use a log e transformation instead of the default log 2.

Normalization produces a Normalized counts data node on the Gene Expression branch of the pipeline (Figure 12). 

 

 

Merge Protein and mRNA data

For quality filtering and normalization, we needed to have the two data types separate as the processing steps were distinct, but for downstream analysis we want to be able to analyze protein and mRNA data together. To bring the two data types back together, we will merge the two normalized counts data nodes.

Data nodes that can be merged with the Antibody Capture branch Normalized counts data node are shown in color (Figure 13).

 

A black outline will appear around the chosen data node. 

The output is a Merged counts data node (Figure 14). This data node will include the normalized counts of our protein and mRNA data. The intersection of cells from the two input data nodes is retained so only cells that passed the quality filter for both protein and mRNA data will be included in the Merged counts data node. 

 

Collapsing tasks to simplify the pipeline

To simplify the appearance of the pipeline, we can group task nodes into a single collapsed task. Here, we will collapse the filtering and normalization steps.

Tasks that can for the beginning and end of the collapsed section of the pipeline are highlighted in purple (Figure 16). We have chosen the Split matrix task as the start and we can choose Merge matrices as the end of the collapsed section. 

 

 

The section of the pipeline that will form the collapsed task is highlighted in green.

 

The new collapsed task, Data processing, appears as a single rectangle on the task graph (Figure 18). 

 

To view the tasks in Data processing, we can expand the collapsed task.

When expanded, the collapsed task is shown as a shaded section of the pipeline with a title bar (Figure 19).

 

To re-collapse the task, you can double click the title bar or click the  icon in the title bar. To remove the collapsed task, you can click the . Please note that this will not remove tasks, just the grouping.

Choosing the number of PCs

In this data set, we have two data types. We can choose to run analysis tasks on one or both of the data types. Here, we will run PCA on only the mRNA data to find the optimal number of PCs for the mRNA data. 

Because we have multiple data types, we can choose which we want to use for the PCA calculation. 

This will generate a Scree plot, which is useful for determining how many PCs to use in downstream analysis tasks. 

A PCA task node will be produced. 

The PCA task report includes the PCA plot, the Scree plot, the component loadings table, and the PC projections table. To switch between these elements, use the buttons in the upper right-hand corner of the task report . Each cell is shown as a dot on the PCA scatter plot. 

The Scree plot lists PCs on the x-axis and the amount of variance explained by each PC on the y-axis, measured in Eigenvalue. The higher the Eigenvalue, the more variance is explained by the PC. Typically, after an initial set of highly informative PCs,  the amount of variance explained by analyzing additional PCs is minimal. By identifying the point where the Scree plot levels off, you can choose an optimal number of PCs to use in downstream analysis steps like graph-based clustering and t-SNE. 

In this data set, a reasonable cut-off could be set anywhere between around 10 and 30 PCs. We will use 15 in downstream steps. 

Cluster by Gene Expression data

CITE-Seq data includes both gene and protein expression information. When the data types are combined, we can perform downstream analysis using both data types. We will begin with the mRNA data.

Once Graph-based clustering has finished running and produced a Clustering result data node, we can visualize the results using UMAP or t-SNE. Both are dimensional reduction techniques that group cells with similar expression into visible clusters. 

The Analyses tab now includes a UMAP task node (Figure 18).

 

The UMAP task report includes a scatter plot with the clustering results coloring the points (Figure 19).

 

An advantage of UMAP over t-SNE is that is preserves more of the global structure of the data. This means that with UMAP, more similar clusters are closer together while dissimilar clusters are further apart. With t-SNE, the relative positions of clusters to each other are often uninformative.  

Classify from expression and clustering

Because we merged the gene and protein expression data, we can visualize a mix of genes and proteins on the gene expression UMAP.

This will color the plot by NKG7 gene expression, a marker for cytotoxic cells. We can color by two T cell protein markers to distinguish cytotoxic T cells from helper T cells. 

This will color the plot by NKG7 gene expression and CD4 protein expression, a marker for helper T cells. We can add a third feature.

This will color the plot by NKG7 gene expression, CD4 protein expression, and CD3 protein expression. Each feature gets a color channel, green, red, or blue. Cells without expression are black and the mix of green, red, and blue is determined by the relative expression of the three genes. Cells expressing both CD4 protein (red) and CD3 protein (blue), but not NKG7 (green) are purple, while cells expressing both NKG7 (green) and CD3 protein (blue) are teal (Figure 25). CD3 is a pan-T cells marker, which helps confirm that this group of clusters is composed of T cells. 

In addition to coloring by the expression of genes and proteins, we can select cells by their expression levels.

By default, any cell that expresses >= 1 normalized count of NKG7 is now selected (Figure 26).

 

Now, any cell that expresses >= 1 normalized count for NKG7 gene and CD3 protein is selected. You can also require that a cell not express a gene or protein.

We have now selected only cells that express >= 1 normalized count for NKG7 gene and CD3 protein, but also have <= 2 normalized count for CD4 protein (Figure 27).

We can classify these cells. Because they express the pan T cell maker, CD3, and the cytotoxic marker, NKG7, but not the helper T cell marker, CD4, we can classify these cells as Cytotoxic T cells. 

To classify the helper T-cells, we can modify the selection criteria. 

We have now selected the CD4 positive, CD3 positive, NKG7 negative helper T cells (Figure 28).

 

We can check the results of our classification.

To return to the full data set, we can clear the filter.

The zoom level will also be reset (Figure 30).

 

In addition to T-cells, we would expect to see B lymphocytes, at least some of which are malignant, in a MALT tumor sample. We can color the plot by expression of a B cell marker to locate these cells on the UMAP plot. 

 There are several clusters that show high levels of CD19 protein expression (Figure 31). We can filter to these cells to examine them more closely.

 

We can use information from the graph-based clustering results to help us find sub-groups within the CD19 protein-expressing cells.

With the help of the Group biomarkers table, we can quickly characterize a few notable sub-groups based on their clusters (Figure 33).

 

Cluster 7, shown in pink, lists IL7R and CD3D, genes typically expressed by T cells, as two of its top biomarkers. Biomarkers are genes or proteins that are expressed highly in a clusters when compared with the other clusters. Therefore, the cells in cluster 7 are likely doublets as they express both B cell (CD19) and T cell (CD3D) markers. 

The biomarkers for clusters 1 and 2 also show an interesting pattern. Cluster 1 lists IGHD as its top biomarker, while cluster 2 lists IGHA1. Both IGHD (Immunoglobulin Heavy Constant Delta) and IGHA1 (Immunoglobulin Heavy Constant Alpha 1) encode classes of the immunoglobulin heavy chain constant region. IGHD is part of IgD, which is expressed by mature B cells, and IGHA1 is part of IgA1, which is produced by plasma cells. We can color the plot by both of these genes to visualize their expression.

This will color the plot by IGHD and IGHA1 (Figure 35).

 

The clusters on the left show expression of IGHA1 while the larger or the two clusters on the right expresses IGHD. We can use the lasso tool to classify these populations.

We can now classify the cluster that expresses IGHD as mature B cells. 

We can visualize our classifications.

To use these classifications in downstream, we can apply the classifications.

This will produce a Classified groups data node. 

Clustering by protein expression

In addition to performing clustering by gene expression data, we can use the protein data for clustering and UMAP visualization. 

Notice that we did not set the number of PCs in this case. If there are fewer than 50 proteins in the data set, all possible PCs will be used by default and, because using all the PCs will capture all of the variance in the data set, this is equivalent to running clustering on the original data. If you data set has more than 50 proteins and you want to run clustering on full data instead of a subset of PCs, simply set the number of PCs to All in the advanced settings.

We can open the UMAP task report to view the clustering result.

UMAP using the protein expression data resolves the cell types we identified earlier on the gene expression UMAP (Figure 39).

 

We can take a closer look at the helper T cell cluster to see if any additional cell types can be found using the protein expression data.

With that, let's take a look at the clustering results from the protein expression data for these cells.

Please note that Graph-based always refers to the most recent graph-based clustering result in the pipeline. 

 

The far-left cluster, cluster 8, has several interesting biomarkers. The top biomarker, is CXCL13, a gene expressed by follicular B helper T cells (Tfh cells). Two of the other biomarkers are PD-1 protein, which promotes self-tolerance and is a target for immunotherapy drugs, and TIGIT, another immunotherapy drug target. 

PD-1 expression is highest in cluster 8 with uniformly strong expression throughout (Figure 41).

 

It is interesting to note that this pattern of PD-1 expression is not easily discernible at the PD-1 gene expression level (PDCD1) (Figure 42). 

 

The Tfh cell marker, CXCL13, is highly and specifically expressed in cluster 8, so we will classify these cells as Tfh (Figure 43). 

 

Classify cells using Scatter plot

An alternative method to clustering and UMAP/t-SNE for classifying cells is using a scatter plot to visualize the expression of key marker genes or proteins. This approach is more effective with CITE-Seq data than gene expression data alone as the protein expression data has a better dynamic range and is less sparse. 

 

Similar to the t-SNE or UMAP scatter plots, each point on the plot is a single cell. The axes are set to features (gene or protein) in the data set by default, but can be set to any attribute or feature. On this plot, we can see that CD3_TotalSeqB is on the x-axis and CD4_TotalSeqB is on the y-axis. We can use our selection and filtering tools to perform a basic classification of CD4 and CD8 T cells. 

This will select any cell with <= 2 normalized count for CD3 protein. Selected cells are shown in bold on the plot and, because we have CD3_TotalSeqB on one of our axes, the cutoff point chosen can be easily evaluated (Figure 46). 

 

The selected CD3+ cells are our T cells. We can filter to these cells prior to performing our classification of CD4 and CD8 T cells sub-types.

Next, we can switch the x-axis to show CD8 protein expression so that we can perform our classification.

The x-axis now shows CD8a protein expression (Figure 47).

 

We can now use a set of filters to select and classify the CD3+ CD4+ CD8- T cells.

This will select the cells in the upper left-hand section of the plot (Figure 48). 

 

We can now select and classify CD3+ CD4- CD8+ T cells using the filters we have already created.

This selects the cells in the lower right-hand section of the plot (Figure 49). 

 

To view our classifications, we can clear the selection and color by classification.

An alternative approach to using the expression threshold filters is to draw a lasso around the population of interest using the lasso tool  and then classify the selected cells.