Pairwise Adjacency

Counts cell-cell interactions based on a previously-generated NN graph and computes interaction scores (enrichments and p-values) for each cell pair. For symmetrical NN definitions, will only return a single value for each pair, but for directed graphs (e.g. KNN) will return values for each "direction" (e.g. "Macrophages -> Tumor" and "Tumor -> Macrophages").

Usage

pairwiseAdjacency(
  object,
  feature,
  nn,
  method = c("hypergeometric", "permutation"),
  drop.self = TRUE,
  n_permutations = 100,
  p.correction = "bonf",
  analyze = "regions",
  ...
)

Arguments

object: A SpatialMap or Region object.
feature: The name of a cellMetadata column (character).
nn: The name of a nearestNeighborGraph from object (character).
method: What test method to calculate adjacency with - permutation or hypergeometric. Hypergeometric is much faster. See the note below for important differences between these methods if you have multiple Regions in your dataset that belong to the same biological sample.
drop.self: If TRUE, ensures that the index cell is not counted as its own interaction, even if such self-interactions are present in the network-graph (NN object). This is recommended to be TRUE in almost all cases.
n_permutations: If method = "permutation", this dictates how many permutations will be performed.
p.correction: p.value correction to apply to enrichment tests within each Region (passed to stats::p.adjust)
analyze: Argument passed to .smapply. See note below on how to use this argument for sample-level analysis.
...: Arguments passed to .smapply

Value

A SpatialMap or Region object with the results of the pairwise adjacency calculations added as a data frame to the misc slot of the specified nn with the name {feature}.adjacency

Note

Credit to Camara lab for the inspiration (https://doi.org/10.1126/sciadv.abc5464).

While the default options for pairwiseAdjacency run this analysis on a Region basis (typically spatially continuous areas of tissue), this can present statistical problems for the calculations performed by compareAdjacency down the line in the case when multiple images (i.e. Regions) were taken from the same patient or tissue sample. Since compareAdjacency makes an assumption of statistical independence for each Region, you may obtain artificially inflated p-values when running this analysis on a Region level if you have this type of experimental design.

Therefore, when you have multiple Regions associated with the same sample, we recommend running this function with analyze = "sample_label", where "sample_label" can be replaced with any value in your object's projectMetadata(). With this option selected, pairwiseAdjacency will sum together cell interaction counts across all Regions in each sample, with the following approach taken for the two methods:

method = "hypergeometric": The background distribution is derived after summing together the counts of each cell type across all Regions in each sample. This is similar to an assumption of spatial contiguity across all regions (i.e. any given cell is equally likely to have appeared in any Region associated with that sample).
method = "permutation": The background distribution is derived by permuting cell labels within each Region (the same approach that is used without aggregation) and then summing permuted interaction counts across all Regions in each sample.

Examples

sm_skin <- load_sm_data("skin")
sm_skin <- spatialNearestNeighbors(sm_skin)
sm_skin <- pairwiseAdjacency(sm_skin, "cell_type", "spatial_knn_5")
df <- .smapply(sm_skin, function(x) {
  nn.misc(NNs(x, "spatial_knn_5"))[["cell_type.adjacency"]]
}, cat.output = "rbind")
head(df)
#>   Region                    feature1        feature2 num.interactions
#> 1  Skin1             B cells (CD20+) B cells (CD20+)                1
#> 2  Skin1      Keratinocytes (PanCK+) B cells (CD20+)                1
#> 3  Skin1       MDSCs (CD14+ HLA-DR-) B cells (CD20+)                1
#> 4  Skin1                Macs (CD68+) B cells (CD20+)                5
#> 5  Skin1 Melanocytes (PanCK+ CD117+) B cells (CD20+)                1
#> 6  Skin1   Monocytes (CD14+ HLA-DR+) B cells (CD20+)                1
#>   interaction.proportion theoretical.proportion log2.enrichment
#> 1           3.988831e-05           8.101760e-08       8.9435150
#> 2           3.988831e-05           3.102974e-05       0.3623144
#> 3           3.988831e-05           1.539334e-06       4.6955875
#> 4           1.994416e-04           1.725675e-05       3.5307334
#> 5           3.988831e-05           2.576360e-05       0.6306320
#> 6           3.988831e-05           2.471037e-06       4.0127776
#>   p.value.enrichment p.value.depletion      p.value possible.hits total.pairs
#> 1       1.031312e-06         0.9999990 2.062624e-06             2    24685993
#> 2       1.832506e-01         0.8167494 3.665013e-01           766    24685993
#> 3       7.075762e-04         0.9992924 1.415152e-03            38    24685993
#> 4       6.102909e-06         0.9999939 1.220582e-05           426    24685993
#> 5       1.371700e-01         0.8628300 2.743400e-01           636    24685993
#> 6       1.813558e-03         0.9981864 3.627116e-03            61    24685993
#>   possible.non.hits draws        p.adj padj.enrichment padj.depletion
#> 1          24685992 25070 0.0004640905    0.0002320452              1
#> 2          24685228 25070 1.0000000000    1.0000000000              1
#> 3          24685956 25070 0.3184092904    0.1592046452              1
#> 4          24685568 25070 0.0027463091    0.0013731545              1
#> 5          24685358 25070 1.0000000000    1.0000000000              1
#> 6          24685933 25070 0.8161011356    0.4080505678              1
#>   log10.padj log10.padj.enrichment log10.padj.depletion
#> 1 3.33339736             3.6344274                    0
#> 2 0.00000000             0.0000000                    0
#> 3 0.49701427             0.7980443                    0
#> 4 2.56125059             2.8622806                    0
#> 5 0.00000000             0.0000000                    0
#> 6 0.08825602             0.3892860                    0

Usage

Arguments

Value

Note

See also

Examples