Build panel for one cluster (beta)

This UNTESTED function finds the best small marker panel for marking a single cluster, using proportion difference as the metric for determining the starting panel.

buildPanel_oneCluster(
  mapDat,
  clustersF,
  medianDat = NA,
  propIn = NA,
  clust = as.character(clustersF[1]),
  subSamp = NA,
  seed = 10,
  maxSize = 20,
  dexCutoff = 0.001,
  topGeneCount = 100
)

Arguments

mapDat

normalized data of the mapping (=reference) data set.

clustersF

cluster calls for each cell.

medianDat

median value for each leaf

propIn

proportions of cells with expression > 1 in each leaf

clust

which cluster to target?

subSamp

number of random nuclei to select from each cluster, EXCEPT the target cluster; set as NA to not subsample

seed

for reproducibility

maxSize

maximum size of marker gene panel

dexCutoff

criteria for stopping: when improvement in fraction of cells properly mapped dips below this value

topGeneCount

number of top genes by proportion to consider

Value

a matrix of the top marker genes for each cluster. Output matrix includes five columns: clust = cluster; panel = ordered genes in the panel for that cluster; onCorrect = fraction of correctly assigned cells in cluster; offCorrect = fraction of cells correctly assigned outside of cluster; dexTotal = additional dex explained by last gene added.