R/05_Hot_cold_spot_analysis_.R
hotspot_analysis.RdApply Getis-Ord Gi statistic to any numeric variable stored in Seurat object and calculate spatial local hot- and cold-spots for that variable
hotspot_analysis(
seurat = NULL,
feature = NULL,
layer = "data",
sim = 999,
padj.thresh = 0.05,
lag = 1,
seed = 347548,
prune = F
)a Seurat object containing Kandinsky data slot(`KanData()`)
character string specifying the name of the feature to use to compute Getis-Ord Gi* statistics
character string indicating which Seurat layer will be considered to search for the variable of interest
number of Monte Carlo simulations to be run for estimating Local Gi* coefficients significance
numeric value indicating the significance threshold to be applied to the adjusted pvalues resulting from Monte Carlo simulations
integer value indicating the extent of cell/spot neighbours to be considered to calculate Getis-Ord Gi* statistics. `lag = 1` indicates that only 1st order neighbours will be considered, while `lag=2` indicates that all neighbours of each 1st order neighbour will be also considered for each spot/cell, and so on. When the average number or 1st order neighbours is limited as in the case of Visium spots (that is, by default, only spots/bins composing the ring immediately surrounding each spot/bin), increasing the lag might help in reducing the variability of the final results. Default value is set to 1.
numeric, random seed for reproducibility. Default is set to 347548
boolean, whether ignoring any spot/cell annotated as hot or cold spot with no neighbouring hot or cold spots, respectively (i.e., singlets)
updated Seurat object with two new columns stored in the meta.data table: `feature_name_GI` and `feature_name_clust`, reporting the local Gi* statistics and the association with a hot- or cold-spot for each spot/cell, respectively
The function will use the `localG_perm()` function from the `spdep` package to calculate, for a numeric variable of interest (e.g. expression level of a gene), the local Getis-Ord Gi* statistic for each spot/cell analysed. The `hotspot` function from the same package wil be then use to define local hot-spots or cold-spots on the basis of the spot/cell Gi statistics. These results should be interpreted as follows: - Spots / cells classified as hot-spots show significantly high values of the variable of interest, as well as their neighbouring spots/cells - Spots / cells classified as cold-spots show significantly low values of the variable of interest, as well as their neighbouring spots/cells For more information about Getis-Ord Gi* statistics implemented in `spdep`, see https://r-spatial.github.io/spdep/reference/localmoran.html