Request to allow colPair to accept the output from `findKNN`

[doliv071]

I think colPair is a convenient place to put a KNN graph to avoid recalculating it if multiple functions will use it. I realize findKNN doesn't return a well defined object, but I think a quick length, name, dimension check should cover most situations. This is what I'm doing now, apologies for not having a more relevant code snippet, my OOP is more or less non-existent.

colPairKnnHelper <- function(sce, type, knn){
    validSCE(sce)
    if(type %in% colPairNames(sce)){
        stop("type already exists in colPairs")
    }
    stopifnot(is.list(knn) && length(knn) == 2 && 
                  identical(names(knn), c("index", "distance")))
    stopifnot(ncol(sce) == nrow(knn$index) && ncol(sce) == nrow(knn$distance))
    colPair(sce, type)  <- sparseMatrix(i = rep(1:nrow(knn$index), 
                                                length.out = length(knn$index)), 
                                        j = as.vector(knn$index), 
                                        x = as.vector(knn$distance), 
                                        dims = c(nrow(knn$index), nrow(knn$index)))
    return(sce)
}

[LTLA]

I'd say a better approach would be to add a small function in BiocNeighbors to convert the findKNN results into a sparse matrix (or a Hits object), which you could then assign directly to colPair(), e.g., colPair(sce, type) <- convertToSparseMatrix(knn).

[doliv071]

Yeah that makes sense. I can submit an issue over there. I can also submit a pull request with the two conversion helpers since it doesn't look like they will need any methods or generics, but I'm not sure where they would go in the code base. I'm mentioning this here because I assume once I submit an issue over there, it will be you who responds lol

[LTLA]

Probably just stick them in a new convert.R file for the time being, I'll take care of moving them if I decide that there's a better place for them to live.