Motivation

Classifying single cells with their cell phenotypes is an important component of the image analysis pipeline, allowing us to assign biological information to specific cells in a segmented image. Unsupervised clustering techniques leverage the increased parameters of multiplexed data to partition cells into diverse phenotypes based on their unique patterns of protein expression.

Here, we offer a set of guidelines to use Leiden-based unsupervised clustering to identify cell phenotypes from multiplexed fluorescence images. These guidelines are based on our pre-built 51-plex biomarker panel. These steps can be performed using the Unsupervised Clustering Extension on the Portal or via SpatialMap in Workbench.

<aside> 📝 Summary of Workflow. For more complete explanation, check out Clustering Major Cell Phenotypes, Sub-clustering CD8+ T cells, and Sub-clustering Macrophages, below.

Choose Biomarkers

• Does this list include biomarkers to identify cell types relevant to the study?
• Does this list avoid nuclear counterstains?
• Does this list avoid functional markers (unless sub-clustering)?
• Does this list avoid markers with low signal-to-noise ratio?

Troubleshooting

Build a UMAP

• From UMAP colored by cluster, are clusters discrete in UMAP space?
• From UMAP colored by cluster, does data appear over-clustered or under-clustered?
• From UMAP colored by region, are clusters distributed across regions?

Troubleshooting

Manually annotate clusters

• Use the following tools to help annotate clusters:
  • UMAP colored by cluster
  • UMAP colored by marker expression
  • Heat map of median marker expression
  • Image overlays of clusters and key marker expression
  • Cluster statistics

Example

Troubleshooting

</aside>

Clustering Major Cell Phenotypes

Sub-clustering CD8+ T Cells

Sub-clustering Macrophages