FlowSOM: Using self‐organizing maps for visualization and interpretation of cytometry data

Abstract

The number of markers measured in both flow and mass cytometry keeps increasing steadily. Although this provides a wealth of information, it becomes infeasible to analyze these datasets manually. When using 2D scatter plots, the number of possible plots increases exponentially with the number of markers and therefore, relevant information that is present in the data might be missed. In this article, we introduce a new visualization technique, called FlowSOM, which analyzes Flow or mass cytometry data using a Self-Organizing Map. Using a two-level clustering and star charts, our algorithm helps to obtain a clear overview of how all markers are behaving on all cells, and to detect subsets that might be missed otherwise. R code is available at https://github.com/SofieVG/FlowSOM and will be made available at Bioconductor. Key termsKey terms: polychromatic flow cytometry; mass cytometry; exploratory data analysis; visualization method; self-organizing map; bioinformatics AT the moment, many flow cytometry experiments are performed with seven colors or more. For mass cytometry experiments, this number is even higher. Analyzing these high-dimensional datasets is not always easy, as traditional gating relies on selection of defined cell populations. It is difficult and time-consuming to keep an overview of how markers are behaving for all these defined cell types. In practice, not all combinations of markers are examined and therefore, valuable information can remain unexamined and unnoticed.A solution to this problem is the use of advanced visualization techniques in which more information is provided than in the traditionally used scatter plots.Examples of new visualization techniques developed specifically for this purpose are Visne (1) and SPADE (2). Whereas Visne will plot all cells in a transformed twodimensional space, SPADE will cluster cells in many groups and visualize the results in a minimal spanning tree. SPADE is, however, quite slow, especially for larger datasets. For both Visne and SPADE, many plots need to be investigated to get a correct annotation of cluster boundaries and cell types.Completely automatic clustering algorithms like flowMeans, SWIFT and others (3-10) are another solution that might be considered. Yet, even when using these algorithms, it is necessary to visualize the results clearly to interpret them correctly. The problems we described before are intrinsic to using scatter plots, so the same problems remain as with traditional gating if these automatic techniques are not combined with new visualization algorithms.A self-organizing map (SOM) is an unsupervised technique for clustering and dimensionality reduction, in which a discretized representation of the input space is trained. This technique has already been used on flow cytometry data by the Flow-

Computational flow cytometry: helping to make sense of high-dimensional immunology data

“…Another clustering based technique to visualize cytometry data is the FlowSOM (flow cytometry data analysis using self-organizing maps) algorithm 52 . FlowSOM uses self-organizing maps (SOM) to simultaneously cluster and visualize cytometry data in a two-dimensional grid of cell type clusters.…”

Leveraging blood and tissue CD4+ T cell heterogeneity at the single cell level to identify mechanisms of disease in rheumatoid arthritis

“…One particularly intriguing application of Seurat is to use single cell transcriptomic data to reconstruct the spatial organization of cells, which has been demonstrated in zebrafish embryos [90••]. Multiple clustering methods have been developed for the analysis of flow cytometry [91,92] and mass cytometry [46,93–97]; a recent comparison of these methods identified FlowSOM [96] and PhenoGraph [97] as the best performers [98]. …”

Algorithmic Tools for Mining High-Dimensional Cytometry Data

“…FlowSOM is a recently released algorithm that provides a similar visualization capacity to SPADE, but requires significantly less computation time and allows multiple parameter visualization on the same MST (21). The decreased algorithmic runtime is achieved by using a self-organizing map (SOM) instead of the hierarchical clustering used in SPADE.…”