R/methods-topMarkers.R

In roryk/bcbioSinglecell: Single-Cell RNA-Seq Utilities

#' Top Markers
#'
#' @name topMarkers
#' @family Clustering Functions
#' @author Michael Steinbaugh
#'
#' @inheritParams general
#' @param n Number of genes per cluster.
#' @param direction Whether to include "`positive`", "`negative`", or "`both`"
#'   directions of association per cluster.
#' @param coding Only include protein coding genes.
#'
#' @seealso
#' - [dplyr::slice()].
#' - [dplyr::top_n()].
#'
#' @return `grouped_df`.
#'
#' @examples
#' # grouped_df ====
#' # Currently only works on `Seurat::FindAllMarkers()` return sanitized with
#' # our `sanitizeMarkers()` function
#' x <- topMarkers(
#'     object = all_markers_small,
#'     n = 2L,
#'     direction = "positive"
#' )
#' head(x)
NULL



# Methods ======================================================================
#' @rdname topMarkers
#' @export
setMethod(
    "topMarkers",
    signature("grouped_df"),
    function(
        object,
        n = 10L,
        direction = c("positive", "negative", "both"),
        coding = FALSE
    ) {
        stopifnot(.isSanitizedMarkers(object))
        assert_is_subset(c("avgLogFC", "padj"), colnames(object))
        assertIsAnImplicitInteger(n)
        direction <- match.arg(direction)
        assert_is_a_bool(coding)

        if (isTRUE(coding)) {
            assert_is_subset("geneBiotype", colnames(object))
            object <- object %>%
                .[.[["geneBiotype"]] == "protein_coding", , drop = FALSE] %>%
                # Remove genes annotated as "predicted"
                .[!grepl("^predicted\\s", .[["description"]]), , drop = FALSE]
        }

        # Subset to positive or negative correlation, if desired ("direction")
        # Note that `avgDiff` has been renamed to `avgLogFC` in Seurat v2.1
        if (direction == "positive") {
            object <- object[object[["avgLogFC"]] > 0L, , drop = FALSE]
        } else if (direction == "negative") {
            object <- object[object[["avgLogFC"]] < 0L, , drop = FALSE]
        }

        object %>%
            # Arrange by adjusted P value
            arrange(!!sym("padj")) %>%
            # Take the top rows by using slice
            dplyr::slice(1L:n)
    }
)