Store S4 objects in data.frame or data.table - r

Question

I'm trying to put complex S4 objects (generated with Seurat package) in data.table (I read that it was not possible to use a list or a data.frame, but I didn't find anything about the compatibility of data.table with S4 objects) depending on the value of one of their attribute with a function.
These objects all come from a bigger object that I called dataset in the function I wrote:
subsets_by_cluster <- function(dataset){
nclust=data.table(cluster_ID=c(rep(NA,length(unique(dataset#active.ident)))))
for (i in length(nclust)){
nclust[i]=dataset[,dataset#active.ident==unique(dataset#active.ident)[i]]
}
return(nclust)}
I was expecting getting a data.table full of S4 objects, with one column with as many rows as number of different #active.ident values (cluster IDs)
But when I run it on my original dataset, I get the error
Error in [<-.data.frame(*tmp*, i, 1, value = new("Seurat", assays = list( : replacement has 2965 rows, data has 1
I also tried to do it manually with this kind of line
nclust[1]=dataset[,dataset#active.ident==unique(dataset#active.ident)[1]]
but it didn't work either, prompting the error :
type 'S4' cannot be coerced to 'logical'
Storing the subset in a variable works perfectly, but I would like my script be able do handle different cluster numbers.
I was thinking about writing the files to read so they can then be read, but it seems far from being a optimal solution.
Do you have suggestions ?

Answer 1

First, creating a simple S4 class (taken from Hadley Wickham's Advanced R)
setClass("Person",
slots = c(
name = "character",
age = "numeric"
)
)
As #John Paul mentions, you can create a few and store them in a list
john <- new("Person", name = "John Smith", age = NA_real_)
jane <- new("Person", name = "Jane Smith", age = NA_integer_)
myPeeps <- list(john, jane)
Printing the list
> myPeeps
[[1]]
An object of class "Person"
Slot "name":
[1] "John Smith"
Slot "age":
[1] NA
[[2]]
An object of class "Person"
Slot "name":
[1] "Jane Smith"
Slot "age":
[1] NA
Since a data.frame is a special type of list and as we see above a list element can be an S4 object, you can store them in a column as well. You just have to use the I() function
size <- 5
propsToMyPeeps <- data.frame(
propsFrom = I(sample(myPeeps, size, replace = TRUE)),
propsValue = sample.int(10, size, replace = TRUE),
propsTo = I(sample(myPeeps, size, replace = TRUE))
)
By default, the print method for data.frame doesn't know how to coerce our Person to a character string so printing the data.frame will cause an error. But if you subset the column, you can see all the objects are there.
> print(propsToMyPeeps$propsTo)
[[1]]
An object of class "Person"
Slot "name":
[1] "Jane Smith"
Slot "age":
[1] NA
[[2]]
An object of class "Person"
Slot "name":
[1] "John Smith"
Slot "age":
[1] NA
[[3]]
An object of class "Person"
Slot "name":
[1] "John Smith"
Slot "age":
[1] NA
[[4]]
An object of class "Person"
Slot "name":
[1] "Jane Smith"
Slot "age":
[1] NA
[[5]]
An object of class "Person"
Slot "name":
[1] "Jane Smith"
Slot "age":
[1] NA

Answer 2

You can do it like this:
library(Seurat)
library(data.table)
data(pbmc_small)
nclust = data.table(cluster_ID=levels(Idents(pbmc_small)))
nclust$data = lapply(nclust$cluster_ID,function(i){
pbmc_small[,Idents(pbmc_small)==i]
})
And they can be accessed:
library(gridExtra)
grid.arrange(grobs=lapply(nclust$data,DimPlot),ncol=3)
cluster_ID data
1: 0 <Seurat>
2: 1 <Seurat>
3: 2 <Seurat>
the error in your code comes with first defining the column to be only NAs,and replacing them one at a time. And, it should be for for(i in 1:nrow(nclust)) instead of for(i in length(nclust))
If you start by defining it as a list of NAs, it works:
subsets_by_cluster <- function(dataset){
lvl = levels(Idents(dataset))
nclust=data.table(
cluster_ID = lvl,
data=replicate(length(lvl),NA,simplify=FALSE)
)
for (i in 1:nrow(nclust)){
nclust$data[[i]]=dataset[,Idents(dataset)==lvl[i]]
}
return(nclust)}
subsets_by_cluster(pbmc_small)
cluster_ID data
1: 0 <Seurat>
2: 1 <Seurat>
3: 2 <Seurat>