Cytometry FAQ
Why flowsom is the only hosted FCS clustering tool, how to set n_clusters, and why automated gating and differential abundance have no hosted replacement.
Retired tools
flow_gate, flow_cluster and flow_differential are retired and can no longer be run. For FCS clustering use flowsom. There is no hosted tool for automated gating or differential abundance; ask Ivy to run those in Python in the chat's code workspace.
These answers cover how to work with the cytometry tools — FlowSOM clustering, and imaging mass cytometry segmentation and spatial analysis.
Some answers describe the intended interface, not the implemented one
See the note on the Cytometry overview. The IMC answers below describe parameters that are not in the current imc_segment and imc_spatial schemas.
Cytometry FAQ
Which clustering algorithm should I choose?
FlowSOM is the only hosted clustering algorithm for FCS data, through the flowsom tool. PhenoGraph, PARC and Leiden are not available as hosted tools; Ivy can run them in Python in the chat's code workspace if you need them.
How many metaclusters should I set for FlowSOM?
Start with the default of 10 metaclusters, or 20 for a typical immunophenotyping panel (15-30 markers). If you see biologically distinct populations merged into one cluster, increase n_clusters. If clusters are split unnecessarily, decrease it. The SOM grid (xdim and ydim, 10 by 10 by default) sets the resolution that the metaclusters are built from.
Can I gate cells automatically or test differences between groups?
Not with a hosted tool: flow_gate and flow_differential are retired. Run flowsom to get per-event cluster labels, then ask Ivy to do gating or group comparisons in the chat's code workspace.
How does IMC cell segmentation work?
The segmentation tools use deep learning models to identify individual cells in multi-channel tissue images. Cellpose uses a generalist model trained on diverse cell morphologies. DeepCell/Mesmer is specifically optimized for nuclear + membrane segmentation in tissue. Both require at least one nuclear marker channel (e.g., Ir191/Ir193 for IMC, DAPI for IF). Membrane markers (e.g., CD45, Na-K-ATPase) improve accuracy by defining cell boundaries but are optional.
What is a spatial interaction score?
The interaction score measures whether two cell types are found near each other more or less often than expected by random chance. A score of 1.0 means the co-occurrence matches random expectation. Scores above 1.0 indicate spatial attraction (cells are neighbors more often than chance). Scores below 1.0 indicate spatial avoidance. Significance is determined by comparing the observed frequency to a permutation-based null distribution.
How many permutations do I need for spatial analysis?
The default of 1,000 permutations is sufficient for most analyses and provides p-values accurate to 0.001. For publication-quality results, increase to 10,000 permutations. For quick exploratory analysis, 100 permutations are enough to identify strong spatial patterns. More permutations increase computation time linearly.
Can I use these tools for spectral flow cytometry data?
Yes. After spectral unmixing (performed in your acquisition software), the resulting data can be exported as FCS files and clustered with flowsom. The tool treats each unmixed parameter as a standard marker channel.
How do I handle tissue autofluorescence in IMC?
IMC does not suffer from autofluorescence in the same way as fluorescence microscopy, since it uses metal-tagged antibodies detected by mass spectrometry. However, some channels may have non-specific signal from tissue components. The normalize parameter in imc_segment applies percentile normalization (1st-99th percentile) per channel to reduce the impact of outlier pixels. For spatial analysis, consider excluding channels with known non-specific binding.
Read next
Was this page helpful?