R/pickDimReduction.R

In netSmooth: Network smoothing for scRNAseq

setGeneric(
    name = "pickDimReduction",
    def = function(x, ...) {
        standardGeneric("pickDimReduction")
    }
)

#' Pick the dimensionality reduction method for a dataset that gives the
#' 2D embedding with the highest entropy
#'
#' @rdname pickDimReduction
#' @aliases pickDimReduction
#' @param x  matrix or SummarizedExperiment object [GENES x SAMPLES]
#' @param flavors    list of dimensionality reduction algorithms to try.
#'                   Currently the options are "pca", "tsne" and "umap"
#' @param is.counts    logical: is exprs count data
#' @return    name of dimensionality reduction method that gives the highest
#'            2d entropy
#' @importFrom SummarizedExperiment assay
#' @examples
#' x <- matrix(rnbinom(60000, size=1, prob = .1), ncol=100)
#' pickDimReduction(x)
#' @export
setMethod("pickDimReduction",
    signature(x='matrix'),
    function(x, flavors=c('pca', 'tsne', 'umap'), is.counts=TRUE) {
        entropies <- sapply(flavors, function(flavor) {
            calc2DEntropy(dimReduce(x, flavor=flavor, is.counts=is.counts))
        })
        return(names(which.max(entropies)))
    }
)

#' @export
#' @importFrom SummarizedExperiment assay
#' @rdname pickDimReduction
setMethod("pickDimReduction",
    signature(x='SummarizedExperiment'),
    function(x) pickDimReduction(assay(x)))



#' @export
#' @rdname pickDimReduction
setMethod("pickDimReduction",
    signature(x='Matrix'),
    function(x, flavors=c('pca', 'tsne', 'umap'), is.counts=TRUE) {
        entropies <- sapply(flavors, function(flavor) {
            calc2DEntropy(dimReduce(x, flavor=flavor, is.counts=is.counts))
        })
        return(names(which.max(entropies)))
    }
)

#' @export
#' @rdname pickDimReduction
setMethod("pickDimReduction",
          signature(x='DelayedMatrix'),
          function(x, flavors=c('pca', 'tsne', 'umap'), is.counts=TRUE) {
            entropies <- sapply(flavors, function(flavor) {
              calc2DEntropy(dimReduce(x, flavor=flavor, is.counts=is.counts))
            })
            return(names(which.max(entropies)))
          }
)