Analysis
Validation & QC
Reading a FlowSOM Tree: Similarity Is Not a Developmental Timeline
The original FlowSOM algorithm separates nodes, tree layout and meta-clusters. Keeping those layers distinct helps prevent an exploratory map from becoming an unsupported biological story.
Start with the legend before following a branch
A tree invites the reader to follow a path. In an exploratory cytometry presentation, that visual habit can turn proximity into a story about how cells develop. Before drawing such a conclusion, ask what the edges were constructed to represent and whether the experiment measured development at all.
How the original method builds its display
Van Gassen and colleagues' 2015 FlowSOM paper in Cytometry Part A describes four stages: data preparation, self-organizing map training, minimal spanning tree construction and meta-clustering. The map uses more nodes than the expected number of cell types to support visualization. The tree connects node representatives according to similarity in the multidimensional input space; meta-clustering groups node centres separately.
Star charts display marker summaries, while pie charts can compare node assignments with manual gates. The evaluations include FlowCAP benchmark datasets and flow and mass cytometry examples. The authors acknowledge that manual gating used for comparison is subjective. Their visual topology does not itself measure a developmental trajectory or establish a diagnostic classification.
Three questions for a figure caption
In a local presentation, use the caption to explain what a node represents, what determines an edge and what assigns the displayed label. Those three statements can prevent a reader from treating a graphical grouping as if it were an independently established cell identity. If a biological interpretation goes beyond them, give its supporting evidence separately.
As a review exercise, hide the population names and ask a colleague to describe only what the map encodes. Then restore the labels and discuss which additional observations support them. This exercise is a CFCM suggestion for clearer scientific communication, not a new validation of the algorithm.
Separate exploratory description from a clinical claim
A well-explained map can be useful without carrying a clinical performance claim. State the purpose of the analysis, retain its inputs and settings, and identify what would require an additional experiment. If a proposed report relies on a diagnostic label, establish the evidence appropriate to that decision rather than relying on the familiarity of a tree-shaped plot.
Daily Learning Pill
Before interpreting a branch, finish the sentence: 'This edge was drawn because…'. If the answer is similarity, do not silently replace it with time or ancestry.
Explore the CFCM cytometry resources and bring one map caption to your next methods review. Original educational interpretation by ChatGPT/Codex; not independent scientific review or clinical validation.