add `ppp` object to `SpatialExperiment` objects

[mjemons]

In order to use any spatstat functionality users have to provide a planar point pattern object ppp. So far both @sgunz and me compute these ppp objects on the fly for our own packages sosta and spatialFDA each time we want to compute something with spatstat.

We both have come up with a solution where we first convert the spe object into a data.frame and then to a ppp object in a two step process for speed (see this)

This is relatively fast but leads to many "unnecessary" computations because we never store these objects. @sgunz and me were thinking for a long time whether we could add the ppp objects to the SpatialExperiment objects, similary as SpatialFeatureExperiment adds sf objects for using spdep library.

Is something like this easily doable or should we keep computing these objects on the fly? I believe offering them would make using spatstat functions much easier for users.

[HelenaLC]

What is keeping you from storing this currently? I gave it a quick shot and it's straightforward-ish (not sure how SFE does it?). The motivation to not store stuff like spatstat/sf objects is that this would bump dependencies, and might not be what everyone needs. Historically I think we try to keep things minimal (using DataFrame, matrix, list, etc.). SFE doesn't live by that– this has it's own benefits, but also comes at a cost: 150+ dependencies (vs. 65 for SPE), and complications for anyone not interested in spatstat/sf-interop for any reason. (Note that the below does not "help" a lot in your case, because you want to keep updating marks, etc., so perhaps the on-the-fly isn't that expensive and still more sensible...)

# deps
library(VisiumIO)
library(spatstat.geom)
library(SpatialExperiment)

# load data
dir <- list.dirs(system.file(
    file.path("extdata", "10xVisium"),
    package = "SpatialExperiment"), 
    recursive = FALSE, full.names = TRUE)
spe <- import(TENxVisiumList(
    sampleFolders = dir, 
    sample_ids = basename(dir), 
    processing = "raw", 
    images = "lowres"))

# make ppp
spatialCoordsNames(spe) <- c("x", "y")
df <- data.frame(spatialCoords(spe))
ow <- owin(range(df$x), range(df$y))
pp <- ppp(x=df$x, y=df$y, checkdup=FALSE, window=ow)

# stash split ppp (joint ppp doesn't match ncol)
pp <- split.ppp(pp, factor(seq(ncol(spe))))
head(spe$pp <- pp, 2)

Point pattern split by factor 

1:
Planar point pattern: 1 point
window: rectangle = [1212, 9881] x [1251, 9991] units

2:
Planar point pattern: 1 point
window: rectangle = [1212, 9881] x [1251, 9991] units

# get back joint ppp
do.call(superimpose, spe$pp)

Marked planar point pattern: 99 points
Multitype, with levels = 
   1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 
32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 
61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 
90 91 92 93 94 95 96 97 98 99
window: rectangle = [1212, 9881] x [1251, 9991] units

[HelenaLC]

Fwiw, if you do find something like the above useful, I think we could certainly add a section in the SPE vignette to show how to store such info (also polygons, perhaps), which would add to Suggests: but not Imports:. But I would be hesitant to extend SPE when SFE already does that (and years of devel went into it...). Docs are one thing, but altering the class definition - potentially copying SFE - should be well-justified imo.