Once cells from an image are identified and classified into phenotypes, spatial analysis can be performed to understand how cell composition, function, and organization are related. One such type of spatial analysis is pairwise analysis. The goal of pairwise analysis is to identify cell-cell interactions, determine whether two cell types are preferentially co-localized, and infer cell-cell communication and tissue function.
<aside> đź’ˇ Cell-Cell interaction methods determine pairwise cell interactions within a tissue region using one of three metrics.
Cells can be considered interacting if they are within physical proximity of each other. We offer a number of approaches to algorithmically determine cell-cell interactions.
In this method, a Voronoi diagram of cells within an image is created by expanding radially outward from cells’ centroids until their boundaries touch.
Animation of how Voronoi diagrams are generated: cell centroids (black) radially expand until cell boundaries are touching. Watch the full animation here.
Cells are considered to be interacting if they share a boundary in the Voronoi diagram. Of the methods offered, this method has the strictest definition of interaction as only cells that are directly adjacent to one another are considered interacting. It is most useful for exploring direct cell-cell contact within a tissue.
Illustration of Delaunay Triangulation method for determining a cell’s (blue) interacting pairs (grey). Adapted from Yoon T.J. et al. J. Chem. Phys., 2018.
However, in some instances the Delaunay Triangulation method can lead to false positive cell-cell interactions. In generating Voronoi cells from cell centroids, cell boundaries will continue to expand until constrained by another boundary. If there are gaps in the tissue or holes from cells that been removed by QC filtering methods, the boundaries of Voronoi cells will expand to fill these voids, leading to unexpected cell interactions. Consequently, this method is best suited for tissues without large gaps, and tissues with relatively few QC filtering issues.
This method is automatically computed for every Explorer version created in the Portal. You can learn more about features available in the Explorer here. This method is also available in SpatialMap on Workbench.
This method indexes all cells within a fixed distance of a center cell to determine interactions.
Illustration of Fixed-radius nearest neighbors method for determining a cell’s (green) interacting pairs (grey).
Manipulating the size of the radius facilitates exploration of overarching structure to hyperlocal microenvironments, and can therefore be useful to capture cell-cell communication across varying distances. This method is available in SpatialMap on Workbench.
To determine cell-cell interactions, this method indexes the “K” nearest cells for any given cell, based on the Euclidean distance between cells’ centroids. Interacting cells are then defined as the “K” nearest neighbors from each cell. The parameter “K” can be tuned to consider more or less neighbors as interacting with a cell.
