MIKAIA Cellular Neighborhood App Explained

The MIKAIA^® Cellular Neighborhood App can yield statistics about the cellular microenvironment, i.e., identify cell niches or cellular communities. This presents a type of spatial analysis of cell populations detected by the FL Cell Analysis App or H&E Cell AI App, as well as those imported from annotation files (e.g., GeoJson). Additionally, the app can be integrated into comprehensive workflows, such as analyzing a 47-plex spatial proteomics scan generated with MACSima™ (by Miltenyi Biotec).

The underlying algorithm centers in on each cell and then collects what cell phenotypes are present in the current center cell’s neighborhood. The neighborhood can either be defined by a fixed radius (e.g., 50 µm), it can be comprised of the k-nearest neighbors, or a combination of both.

The Cellular Neighborhood App then can do two things:

1. Yield statistics about the average neighborhood composition by cell type and ROI.
2. Re-classify each cell based on its cellular microenvironment, i.e., identify cell niches or cellular communities.

In this app note, these analyses and the resulting diagrams are explained in detail, based on two artificial examples.

Example 1

The following screenshot shows a field of view that comprises three cell types (“class A”/yellow, “class B”/blue, “class C”/orange).
The cells are arranged in four zones, from top to bottom:

1. Yellow & blue mixed (top)
2. Blue only (2^nd from top)
3. Orange & blue mixed (3^rd from top)
4. Orange cells only (bottom)

Example 1 — Neighborhood composition

The following diagram shows the result of this analysis, when defining a neighborhood of max. 5 cells within at most 30 µm radius. It shows for each cell type (A, B, and C) a distance histogram (configured to 3 bins á 10 µm) that states the relative neighborhood cell composition in 3 bins of 10 µm each. Since the absolute area of each distance bin increases, the composition is displayed as a normalized stacked bar chart. It states both the absolute and relative average cell abundance.

Example 1 — Identification of cellular neighborhoods

Since “Cluster” with k=4 was enabled, 4 new cellular neighborhood annotation classes “CN-1” — “CN-4” were created and each cell was assigned to one of these classes (the original cell classes remain unchanged, but are hidden). The clustering algorithm (k-means) identified the four different zones, described above.

Additional diagrams summarize findings about these cellular neighborhoods. The “Abundance by ROI” simply states the abundance per type and ROI. In this simple example, the analyzed region was not divided into multiple regions of interests.

The two “Cluster Explorer” diagrams give insights on how the individual cellular neighborhoods (CN) are composed. The following diagram shows a 2D heatmap, where each row relates to a CN and each column to one of the original cell types.

From it, we learn how the 4 CNs are composed on average. The average cell abundances per row sum up to 6, since they consider the center cell + the 5 neighbors:

• CN-3: Equal distribution of yellow and blue cells. The pattern consists on average of 3,0 yellow cells and 3,0 blue cells, and is found in the topmost horizontal zone of the analyzed field of view (FoV).
• CN-1: This pattern consists mainly of blue cells with a small amount of orange cells and yellow cells. The pattern consists on average of 5,1 blue cells, 0,2 yellow cells, and 0,7 orange cells. It is found in the horizontal zone 2^nd from the top in the analyzed FoV.
• CN2: Almost equal distribution of blue and orange cells. The pattern consists on average of 2,4 blue cells and 3,6 orange cells. It is found in 3^rd zone from the top.
• CN-4: This pattern consists of mainly orange cells: on average 5.9 orange cells and only 0.1 blue cells. It is found in the bottommost horizontal zone of the analyzed FoV.

The “Cluster Explorer (100%)” visualizes the relative compositions using horizontally stacked bars and additionally states the absolute abundances.

Example 2

Here is another artificial example that is simplified to the point that the absolute and relative cell abundances reported in the diagrams can be easily related to the image content.

Example 2 — Neighborhood composition

Like in the first example, the distance diagrams by cell type again show cell compositions at three distances (0-10 µm, 10-20µm, and 20-30µm).

• class A cells only have one orange neighbor on average (at a distance between 10 and 20 µm). class B cells also have only orange neighbors — in one of the three instances (0,33), the orange cell is within 10 µm and in two cases (0,67) they are located at a distance between 10 and 20 µm.
• Cells of class C have both blue and yellow neighbors: on average 0.5 blue and 0.5 yellow ones. The fact that half of the class C cells only have blue neighbors and the other half only have yellow neighbors cannot be derived from this diagram, but only from the cluster explorer (below).
• Finally, cells of class D have only neighbors that are also of class D — on average 0,29+3,71+2=6, which corresponds to the 3 groups of 7 green cells each in the screenshot above.

Example 2 — Identification of cellular neighborhoods

In this example, the analysis was configured to assign cells into k=3 cellular neighborhood types.

These patterns were found:

• CN-1: on average, 7 cells from class D
• CN-2: on average, 2 cells: 1x class A + 1x class C
• CN-3: on average 2 cells: 1x class B + 1x class C

---

This app note was created with MIKAIA^® v2.3. Learn more about MIKAIA^® apps and AIs.