Vectorized solution to convert UCell scoring output to labels, dealing with ambiguous calls.

convertUCell.R

#' Take output of UCell::ScoreSignatures_UCell() and return a vector of labels with highest score.
#' If the highest score per row is unique (only one value is the highest value) return that label.
#' If all scores are zero return "NA".
#' If more than one value is the highest value return "ambiguous".
#' library(UCell)
#' data(sample.matrix)
#' gene.sets <- list(Tcell_signature = c("CD2","CD3E","CD3D"),
#'                   Myeloid_signature = c("SPI1","FCER1G","CSF1R"))
#' scores <- ScoreSignatures_UCell(sample.matrix, features=gene.sets)
#' convertUCell(scores)
#'
convertUCell <- function(ucell){
  
  isRowMax <- Biobase::rowMax(ucell)
  
  # find cells with only zeros
  j.allZero <- names(which(rowSums(ucell==0)==ncol(ucell)))
  
  # find cells that have > 1 entry with identical but non-zero score
  j.ambiguous <- setdiff(names(which(rowSums(ucell==isRowMax) > 1)), j.allZero)
  
  # the rest is unique
  j.unique <- setdiff(rownames(ucell), c(j.allZero, j.ambiguous))
  u.unique <- ucell[j.unique,]
  l.unique <- colnames(u.unique)[as.numeric(apply(u.unique, 1, which.max))]
  
  # put together
  d <- data.frame(label="NA", names=rownames(ucell), row.names=rownames(ucell))
  
  d[j.allZero,"label"] <- "NA"
  d[j.ambiguous,"label"] <- "ambiguous"
  d[j.unique,"label"] <- l.unique
  
  gsub("_UCell", "", d[rownames(ucell),"label",drop=TRUE])
  
}