evaluateDimRed {pipeComp} | R Documentation |
Gathers evaluation statistics on a reduced space using known cell labels. If using 'evaluteDimRed' in a custom pipeline, you will probably want to use 'pipeComp:::.aggregateDR' as the corresponding aggregation function.
evaluateDimRed(x, clusters = NULL, n = c(10, 20, 50), covars)
x |
The matrix of the reduced space, with cells as rows and components as columns |
clusters |
The vector indicating each cell's cluster. |
n |
A numeric vector indiciating the number of top dimensions at which to gather statistics (default 'c(10,20,50)'). Will use all available dimensions if a higher number is given. |
covars |
A character vectors containing any additional covariates (column names of 'colData') to track during evalutation. If missing, will attempt to use default covariates. To disable, set 'covars=c()'. |
A list with the following components: * silhouettes: a matrix of the silhouette for each cell-cluster pair at each value of 'n' * clust.avg.silwidth: a matrix of the cluster average width at each value of 'n' * R2: the proportion of variance in each component (up to 'max(n)') that is explained by the clusters (i.e. R-squared of a linear model).
# random data library(scater) sce <- runPCA(logNormCounts(mockSCE(ngenes = 500))) sce <- addPerCellQC(sce) # random population labels sce$cluster <- sample(LETTERS[1:3], ncol(sce), replace=TRUE) res <- evaluateDimRed(sce, sce$cluster, covars=c("sum","detected")) # average silhouette widths: res$clust.avg.silwidth # adjusted R2 of covariates: res$covar.adjR2