Expression Hotspots (Getis-Ord Gi*)

Answers the question: Where does a selected gene form spatial clusters of high or low expression?

For a selected gene, the tool considers each cell together with its $k$ nearest neighbours and calculates a Getis-Ord Gi* z-score. This score measures whether gene expression in that local neighbourhood is higher or lower than expected if the observed expression values were randomly distributed across the cells.

More specifically, the Gi* score expresses the difference between the observed and expected neighbourhood expression in units of the variation expected under random mixing:

Here, $N_i$ contains cell $i$ and its $k$ nearest neighbours, $m_i$ is its size, $x_j$ is the selected gene’s expression in cell $j$, and $\overline{x}$ and $S$ are its mean and standard deviation across all $n$ mapped cells.

Map. Each cell is colored by its Gi* z-score. Red shows locally elevated expression, while blue shows locally reduced expression. Because Gi* includes the cell itself, the map describes the expression level of the whole local region rather than whether one cell differs from its neighbours.

Summary. Gi* z-scores are converted into two-sided p-values and corrected across cells using the Benjamini–Hochberg method by default. Click the info icon after the run to see how many cells were classified as significant hotspots or coldspots at the selected q-value threshold.

Table. Each row represents one cell type. In hotspot and In coldspot count its cells in significant regions. Hotspot fraction gives the proportion of that cell type found in hotspots, while Mean Gi* shows its average hotspot score.