Flexible grouping of rows/columns of a raw adjacency matrix

Source: R/hemibrain_adjacency.R

Create a summary adjacency matrix in which input and/or output neurons are coalesced (by summation) into a smaller number of rows/columns. This can be used as a preprocessing step to present a simple representation of the connectivity between a larger number of cells in a heatmap.

grouped_adjacency(
  inputids = NULL,
  outputids = NULL,
  ingroup = c("type", "name", "cellBodyFiber"),
  outgroup = c("type", "name", "cellBodyFiber"),
  threshold = c(1, 0),
  scale = c("none", "col", "row"),
  ...
)

Arguments

inputids

Either the bodyids of the input neurons OR an adjacency matrix
outputids

bodyids of the output neurons. Not required if inputids is an adjacency matrix.
ingroup, outgroup

grouping variables used to coalesce related neurons. The default values match against dynamically fetched neuprint metadata columns for the neurons specified by inputids,outputids.You can also specify a factor of the appropriate length or NULL to suppress grouping. See details and examples.
threshold

Remove (groups of) neurons that make fewer connections than this. If you provide a length two vector the first threshold will be an absolute value applied before scaling, while the second will be a fractional value applied after scaling.
scale

Whether to scale the rows or columns so that they sum to 1. Default is 'none'.
...

Additional arguments passed to neuprint_get_adjacency_matrix

Value

named numeric matrix with rows for input groups and columns for output groups.

Details

The default value of threshold is c(1, 0). When scale=F this will not apply a threshold. When scale=T this will apply a threshold of 1 to the raw counts before scaling to ensure that only valid output will be returned.

Grouping. There are 3 ways to specify ingroup, outgroup grouping variable:

  • by a string chosen from c("type", "name", "cellBodyFiber") which specifies a neuprint field; this is retrieved dynamically for the body ids specified by inputids, outputids.

  • by a factor or numeric vector specifying groups. If the elements in this vector are named, then the groups

  • by a function that maps body ids to numeric/factor groups that are then handled like the previous option.

See also

neuprint_get_adjacency_matrix

Examples

# \donttest{
# Default: search by type and group by type
# NB first search is by regex
# do not group input neurons but leave one row for each neuron
da2pnkc=grouped_adjacency("/.*DA2.*PN.*", 'KC', ingroup = NULL)
heatmap(da2pnkc)


# alternatively, if you want to play around with different arguments,
# you can get the raw adjacency matrix and then group that in different
# ways
if (FALSE) {
pnkc.raw=neuprint_get_adjacency_matrix(inputids = class2ids("PN"),
  outputids = 'KC')
pnkc.bytype=grouped_adjacency(pnkc.raw, ingroup = "type", outgroup = "type")
heatmap(pnkc.bytype)

# slightly finer groups
heatmap(grouped_adjacency(pnkc.raw, ingroup = "type", outgroup = "name"))
# add a threshold to remove neurons that make very few connections
heatmap(grouped_adjacency(pnkc.raw, ingroup = "type", outgroup = "name",
  threshold=50))

# Trying scaling to see relative input onto different groups
# of Kenyon cell target neurons. col=>
heatmap(grouped_adjacency(pnkc.raw, ingroup = "type", outgroup = "name",
  scale='col', threshold = c(1,0.01)))
}
# }