FlowSOM, SPADE, and CITRUS on viSNE: automatically categorize viSNE populations

Table of Contents


viSNE is an excellent method for reducing high dimensional data to two dimensions and thereby enabling rapid exploratory data analysis and visualization of complex results. For cytometry data, this may assist with the categorization of events/cells into biological populations. For bulk data, this may help you understand the heterogeneity in your samples. In either case, it is sometimes useful to categorize groups seen on the viSNE map for downstream analysis. This can be done using gates:


(areas of a viSNE map categorized into populations using gates)


To learn more about this process, read our article about gating a viSNE map. Unfortunately, gating a viSNE map can be a time-consuming, subjective and detail-driven process. An alternative to this is to use a computer-driven clustering method to categorize groups seen on the viSNE map automatically. There are multiple approaches that can be used for clustering a viSNE map, one of which is running FlowSOM, SPADE, or CITRUS on the coordinates of the viSNE map itself. The end result of running FlowSOM, SPADE, or CITRUS on a viSNE map is a collection of clusters that correspond to spatial locations on the viSNE map. Another approach that leverages the display of clustering data on a viSNE map is to run SPADE, FlowSOM, or CITRUS on the population-defining markers (e.g. CD markers) and to display the resulting clustering data overlaid on the viSNE map. This can facilitate the quality assessment and iteration of algorithm run settings, and provides a starting point for analysis of clustering data.


When to run a clustering algorithm on dimensionality reduction (e.g. viSNE) channels


Clustering on dimensionality reduction channels (e.g. viSNE channels) can be a useful approach for defining groups of cells or groups of samples when the dimensionality of your data is very high. In these cases, the "curse of dimensionality" may cause a clustering method to be unable to perform well unless you first reduce the dimensionality of the data. Unfortunately, because every dataset is different, it's hard to know when you may reach this point. If your data are very high dimensional cytometry data, data with hundreds of markers measured in all of your samples, or you are noticing that your clustering results don't make sense, clustering on dimensionality reduction (e.g. viSNE) channels may be a better option for defining groups of cells or groups of samples.


When to Display Clusters (e.g. from FlowSOM/SPADE/CITRUS) on Dimensionality Reduction Maps (e.g. viSNE)


If clustering on viSNE channels isn’t appropriate for your workflow, you may still benefit from combining clustering and viSNE data alongside your native data to allow you to display clustering data on a viSNE map. This can help with clustering algorithm optimization and with assessment of cluster identity. In this workflow, you cluster on the typical population-defining channels (e.g. CD markers), gate the clusters, and overlay the gated cluster Populations on the viSNE map. You can run viSNE either before or after running the clustering algorithm.


Directions for Running a Clustering Algorithm on viSNE Data in Cytobank

1) Navigate to the viSNE analysis

In order to run a clustering algorithm on viSNE results, first navigate to a viSNE experiment (see How to Configure and Run a viSNE Analysis).

2) Choose a clustering method

Within the viSNE experiment that houses the viSNE result files, choose a clustering algorithm to configure within the Advanced Analyses menu. Options include FlowSOM, SPADE, CITRUS, and viSNE.

Note: Cytobank no longer requires you to clone a viSNE experiment before running a clustering algorithm on the viSNE results. You can run a new algorithm directly within the viSNE experiment.


The following articles provide detailed information on how to configure a clustering analysis:



Here are some guidelines that apply specifically to this workflow of clustering on viSNE channels:




The files being included for this clustering analysis are the results from your previous viSNE analysis. Thus, the Ungated population actually corresponds to the population that was previously chosen for viSNE. For that reason, simply choose ungated for the clustering analysis. A more restrictive population can be chosen if desired for some other workflow objective.


Clustering Channels


For the workflow of clustering on the viSNE channels to overcome the “curse of dimensionality”:

Choose only the two tSNE channels. This application is for clustering the tSNE map only and thus other channels should not be included for the clustering step.


For the workflow of displaying clustering data on a viSNE map via a standard clustering approach:

Choose only the population-defining markers (e.g. CD markers). The other markers will carry through to the analysis even if they aren’t selected as Clustering Channels.


Fold-Change Groups (applicable to SPADE only)


The typical logic applies for choosing fold change groups and baselines with SPADE. This is a useful way of getting fold change visualizations for a viSNE map, which is usually not possible due to the single cell nature of viSNE results.


Number of Clusters (Nodes)


The number of clusters may need to be honed empirically, but a good starting place may be ~7 times the number of populations you expect to find based on manual gating. On some of the publicly available datasets published in Weber & Robinson (2016), we demonstrated that starting with a number of clusters equal to ~7 times the number of expected populations based on manual gating, we were able to capture all of the populations with a frequency > 0.5% with an F measure that was comparable to the other clustering methods used in that paper.


Downsampled Events Target (applicable to SPADE only)


Running SPADE on viSNE does not require downsampling to aid algorithm performance in the same way that SPADE on high parameter cytometry data requires it. Therefore, set the target to 100 percent. This means that all of the events will be included.

Analysis of the Results of Running a Clustering Algorithm on viSNE Data in Cytobank


One approach to analyzing the results of clustering on viSNE channels (tSNE1 and tSNE2 selected as the Clustering Channels) is to proceed with a typical analysis workflow. For SPADE, this means the typical analysis of a normal SPADE run, including coloring by channel, bubbling, fold change analysis, statistics, exporting FCS files based on bubbles, etc. The way in which the SPADE tree was created is different, but the analysis follows the same principles. The same is true for FlowSOM and CITRUS output.


Regardless of whether you chose to cluster on the viSNE channels or on the population-defining markers, it can also be informative to visualize the clusters overlaid on a viSNE map. This approach can help you gauge the quality of the clustering and can aid in refinement of clustering algorithm settings. Generally, the colors of the overlaid clusters should correspond to the viSNE continent definitions. If the same cluster spans multiple continents, that might indicate a need to increase the target number of clusters (or clusters and metaclusters, for the case of overlaying FlowSOM metaclusters on a viSNE map), or a need to enable or disable normalization. You can also use the Automatic cluster gates functionality to create Populations out of clusters or metaclusters, or to refine the results of FlowSOM-driven metaclustering.


You can compare overlaid clusters on a viSNE map to a contour plot of the viSNE map to gauge whether the number of clusters generated was sufficient to capture distinct, dense populations. In the example above, you can see how a FlowSOM target number of metaclusters of 4 does not provide enough resolution, and a target of 15 provides too much resolution (though you could further explore this to see if, in fact, it is detecting relevant sub-populations). A target number of metaclusters of 7 appears to yield good correspondence between the FlowSOM metaclustering and the viSNE continents (note that in this dataset, the central, pink continent is identifiable only as being CD16+ based on the staining reagents used in the experiment; hence, the identification of one metacluster despite the two dense regions visible in the contour plot view). When a FlowSOM run finishes, you land automatically in the Illustration Editor configured with this view of FlowSOM metaclusters on a viSNE map if your dataset also has the tSNE1 and tSNE2 channels, making it really easy to perform this quality assessment rapidly.


This is same dataset passed through FlowSOM with all settings identical, with the exception of Normalization now being enabled (it was disabled in the prior example):


While normalization is often a helpful and necessary approach to transforming channel data prior to analysis, for some datasets it is not (see example above), and that can be rapidly detected via the display of FlowSOM metaclusters on a viSNE map, followed up by inspection of cluster expression:


If you find that enabling normalization appears to diminish the resolution of the metaclustering, you can also try increasing the target number of metaclusters. There may be cases where enabling normalization appears to lower the resolution and result in poor quality, but by increasing the target number of metaclusters, you may be able to pull out sub-populations.


Now that you have optimized your clustering algorithm run, you can now proceed to view the expression of markers of interest in heatmap format, or to drill down into single-cell views.


How to perform the analysis workflow with FlowSOM:


When a FlowSOM run completes, it writes FlowSOM_cluster_id and FlowSOM_metacluster_id channels into the newly generated files that comprise the FlowSOM analysis experiment. If you have followed the steps above and run viSNE on the files first, the files in the FlowSOM analysis experiment will now contain all of the original channels and data, as well as the tSNE1 and tSNE2 channels from the viSNE run, and the new FlowSOM_cluster_id and FlowSOM_metacluster_id channels. The illustration will be configured by default to set up dot plots with metaclusters overlaid on the viSNE map. (You can also run FlowSOM first, and then run viSNE - the same smart  illustration view will be set up following the viSNE run.) From here, you can assess metacluster identity by setting up heatmaps or channel-colored dot plots to view the expression of clusters of interest.


Completing the analysis workflow with SPADE:


First, run viSNE on your dataset. Then run SPADE per the instructions above. Following completion of the SPADE run, start in the SPADE result (tree viewer page) and draw a single bubble around the entire SPADE tree. Next, export the bubble as new FCS files. The resulting files will have a Cluster ID channel that can be used to draw cluster gates, which then will allow the visualization of the viSNE map colored by cluster using colored overlay populations, or heatmap display of cluster expression for markers of interest. You’ll set the x- and y-axes to the tSNE1 and tSNE2 channels, select your cluster gates for the Populations, and set the plot type to Dot colored by Overlaid Figure Dimension.


Completing the analysis workflow with CITRUS:


First, run viSNE on your dataset. Then run CITRUS per the instructions above. Following completion of the CITRUS run, export the clusters in which you are interested to a new experiment. Set the x- and y-axes to the tSNE1 and tSNE2 channels, and ensure your files are annotated, leveraging the Figure Dimensions. For example, if you have used the Sample Types figure dimension to annotate your CITRUS cluster files, put the Sample Types Figure Dimension in the second position and set the plot type to Dot colored by Overlaid Figure Dimension.

Have more questions? Submit a request