pcantalupo/score_table_functions.R

## score_table_functions.R
# Aggregate a score table from single cell level to the cluster level
# Params
#   scores - data.frame or matrix where rows are cells and columns are celltypes.
#            Rownames or colnames are not required
#   clusters - vector of same length as number of rows in 'scores'.
#              The values specify the group for each row (i.e. c(3, 1, 0, 1, 2, 0, 3, ...))
# Return - A matrix with the mean scores. Columns are in the same order.
#          Rows are sorted based on the typeof(clusters) (integer -> integer sorted, character -> character sorted).
#          Rownames are set to the cluster value
scores_clusterlevel = function(scores, clusters) {
  data = apply(scores, 2, function(row) {
    aggregate(row, list(Clusters = clusters), mean)$x   # only need 'x' where 'x' is the mean
  })
  rownames(data) = as.character(sort(unique(clusters)))
  return(data)
}

# Determine the label for each row based on maximum value of the celltype scores in the row.
# scores - rows are cells or clusters and columns are celltypes
#          Rownames and colnames are required.
# unknown - do you want to calculated Unknown labels? (if TRUE, must supply 'clusters' param)
#         - only supported for cluster scores table
# clusters - vector that specifies the cluster for each cell (only needed if calculating unknown labels)
#          - only supported for cluster scores table
# cutoff - if the score is less than this value, the label is 'Unknown' (default is 0.25; based on ScType)
#          - only supported for clusters scores table
# Return - The scores table with 3 preprended columns 'clusters', 'labels', and 'score'
#          'clusters' values are taken from the rownames
#          'labels' the celltype with the maximum score
#          'score' the maximum score
add_labels_based_on_max = function(scores, unknown = FALSE, clusters, cutoff = 0.25) {
  whichmax = apply(scores, 1, which.max)  # for Group 0 the max value is column 3
  labels = colnames(scores)[whichmax]  # extract the label for each Group
  score = sapply(seq_along(whichmax), function(i){  # extract the score for each Group
    maxindex = whichmax[i]
    scores[i,maxindex]
  })
  data = data.frame(clusters = rownames(scores), labels = labels, score = score, scores)
  if(unknown) {
    #numcells = table(clusters)
    data$labels[data$score < cutoff] = "Unknown"
  }
  return(data)
}

# Purpose: You have cluster labels (in your cluster scores table) and you want to expand these labels to the cell level so you can add them to a table of cell metadata
# scores - data.frame of cluster scores containing at least two columns: 'clusters' and 'labels' (rows are clusters)
# clusters - vector that specifies the cluster for each cell
deconvolute_clusterlabel = function(scores, clusters) {
  return(scores$labels[match(clusters, scores$clusters)])
}
	# Aggregate a score table from single cell level to the cluster level
	# Params
	# scores - data.frame or matrix where rows are cells and columns are celltypes.
	# Rownames or colnames are not required
	# clusters - vector of same length as number of rows in 'scores'.
	# The values specify the group for each row (i.e. c(3, 1, 0, 1, 2, 0, 3, ...))
	# Return - A matrix with the mean scores. Columns are in the same order.
	# Rows are sorted based on the typeof(clusters) (integer -> integer sorted, character -> character sorted).
	# Rownames are set to the cluster value
	scores_clusterlevel = function(scores, clusters) {
	data = apply(scores, 2, function(row) {
	aggregate(row, list(Clusters = clusters), mean)$x # only need 'x' where 'x' is the mean
	})
	rownames(data) = as.character(sort(unique(clusters)))
	return(data)
	}

	# Determine the label for each row based on maximum value of the celltype scores in the row.
	# scores - rows are cells or clusters and columns are celltypes
	# Rownames and colnames are required.
	# unknown - do you want to calculated Unknown labels? (if TRUE, must supply 'clusters' param)
	# - only supported for cluster scores table
	# clusters - vector that specifies the cluster for each cell (only needed if calculating unknown labels)
	# - only supported for cluster scores table
	# cutoff - if the score is less than this value, the label is 'Unknown' (default is 0.25; based on ScType)
	# - only supported for clusters scores table
	# Return - The scores table with 3 preprended columns 'clusters', 'labels', and 'score'
	# 'clusters' values are taken from the rownames
	# 'labels' the celltype with the maximum score
	# 'score' the maximum score
	add_labels_based_on_max = function(scores, unknown = FALSE, clusters, cutoff = 0.25) {
	whichmax = apply(scores, 1, which.max) # for Group 0 the max value is column 3
	labels = colnames(scores)[whichmax] # extract the label for each Group
	score = sapply(seq_along(whichmax), function(i){ # extract the score for each Group
	maxindex = whichmax[i]
	scores[i,maxindex]
	})
	data = data.frame(clusters = rownames(scores), labels = labels, score = score, scores)
	if(unknown) {
	#numcells = table(clusters)
	data$labels[data$score < cutoff] = "Unknown"
	}
	return(data)
	}

	# Purpose: You have cluster labels (in your cluster scores table) and you want to expand these labels to the cell level so you can add them to a table of cell metadata
	# scores - data.frame of cluster scores containing at least two columns: 'clusters' and 'labels' (rows are clusters)
	# clusters - vector that specifies the cluster for each cell
	deconvolute_clusterlabel = function(scores, clusters) {
	return(scores$labels[match(clusters, scores$clusters)])
	}