overcluster {scDblFinder} | R Documentation |
This function deliberately overclusters based on the desired range of cluster size. It first calculates a SNN network viar 'scran::buildSNNGraph', then runs 'igraph::cluster_fast_greedy' until no cluster is above the size limits, and merges clusters that are too small. By default, 'rankTrans' is used on the counts before, because it tends to produce over-clustering influenced by library size, which is desirable for producing artificial doublets.
overcluster(x, rtrans = c("rankTrans", "scran", "none"), min.size = 50, max.size = NULL)
x |
A numeric matrix, with entities (e.g. cells) as columns and features (e.g. genes) as rows. Alternatively, an object of class 'igraph'. |
rtrans |
Transformation to apply, either 'rankTrans' (default, dense step-preserving rank transformation, see ‘rankTrans'), ’scran' (default; see ‘scran::scaledColRanks'), or ’none' (data taken as-is). Ignored if 'x' is an 'igraph'. |
min.size |
The minimum cluster size (applies after splitting, and hence overrides 'max.size') |
max.size |
The maximum cluster size. If omitted, will be calculated on the basis of the population size and initial number of clusters. |
A vector of cluster labels.
m <- t(sapply( seq(from=0, to=5, length.out=50), FUN=function(x) rpois(50,x) ) ) cc <- suppressWarnings(overcluster(m,min.size=5)) table(cc)