How are cell segmentations represented within Seurat? (Space Ranger 4.0 support)

[anashen]

Originally posted by @asmlgkj at #10028 (comment).

---

I am working with Seurat v5 and cell segmentation data (polygon boundaries) in your tutorial data. I would like to clarify how Seurat defines and stores the “location” of each segmented cell.

Where exactly in the Seurat object can I find the coordinates that represent each cell’s position?

For polygon-based cells, does Seurat use the geometric centroid of the polygon, or some other representative point?

If it is using the centroid, how does Seurat handle cases where the centroid falls outside of the polygon — for example, in crescent-shaped or very elongated/narrow cells? Is there any correction (such as using an interior point), or does Seurat simply return the centroid as is?

I would be very grateful if you could clarify this, since it has important implications for downstream analyses and visualization.

---

[anashen]

@asmlgkj -- hope this helps! Let me know if you have further questions.

Where exactly in the Seurat object can I find the coordinates that represent each cell’s position?

For centroids, you can retrieve the coordinates with GetTissueCoordinates(v3d). Within the Seurat object you will find coordinates at: "images" slot -> Visium object ("slice1.polygons") -> boundaries -> Centroids object ("centroids") -> "coords" slot. See the screenshot below.

For segmentations, you can retrieve the segmentation coordinates with GetTissueCoordinates(v3d[["slice1.polygons"]]@boundaries[["segmentation"]]). Here, each pair of coordinates corresponding to a single cell ID represents one vertex of a cell segmentation, so you will see multiple pairs of coordinates attributed to one cell ID (thus making the data frame of coordinates very large).

More info: coordinates are stored inside each POLYGON object in thesf object "sf.data". Within the Seurat object you will find polygons at: "images" slot -> Visium object ("slice1.polygons") -> boundaries -> Segmentation object ("segmentation") -> sf.data -> "geometry" column. See the screenshot below (note that the highlight is showing how coordinates are stored for a single cell segmentation).

For polygon-based cells, does Seurat use the geometric centroid of the polygon, or some other representative point?

We use st_centroid from the sf library to return the geometric centroid of each polygon / cell segmentation.

If it is using the centroid, how does Seurat handle cases where the centroid falls outside of the polygon — for example, in crescent-shaped or very elongated/narrow cells? Is there any correction (such as using an interior point), or does Seurat simply return the centroid as is?

For now, we return the centroid as-is for every cell segmentation, which is the fastest method of deriving point representations of the segmentations. However, please note that this is the initial version of support for cell segmentation data, so we may revise this at a later date if users raise issues with the current method.

[asmlgkj]

thanks a lot , Regarding centroid coordinates for downstream analysis: Could you confirm that obtaining centroid coordinates for downstream analysis is indeed feasible? Also, am I correct in understanding that we should use nucleus_segmentations.geojson rather than cell_segmentations.geojson to obtain centroid coordinates? I assume this is because the nucleus is definitively located within the cell, making it more accurate?