Builds a low-dimensional autofluorescence basis from an unstained control as the leading right singular vectors of the raw, uncentred unstained matrix. The leading component is therefore the mean background direction and the remainder describe how background varies from cell to cell.
A single autofluorescence spectrum cannot represent that variation. In a real sample autofluorescence is a family of profiles that tracks cell type, and cell type also determines which markers are expressed, so whatever a single row leaves behind is correlated with marker expression and is readily mistaken for spillover by any method that fits one channel against another.
Usage
get.af.basis(
unstained,
n.pc = "auto",
max.pc = 6L,
read.var = 125^2,
n.permutations = 3L,
max.events = 50000L,
verbose = TRUE
)Arguments
- unstained
Numeric matrix (events x detectors), raw unstained control, already gated to the population of interest.
- n.pc
Integer or
"auto". Number of components to retain."auto"keeps every component whose singular value exceeds the largest value a matrix of the same shape containing only read noise would produce. Default"auto".- max.pc
Integer, hard cap on the number of components. Default
6.- read.var
Numeric, per-detector read noise variance in raw units, the floor of the retention threshold when
n.pc = "auto". Default125^2.- n.permutations
Integer, permutations used to build the retention threshold when
n.pc = "auto". Each permutation costs one decomposition. Default3.- max.events
Integer, maximum events used for the decomposition. Default
50000.- verbose
Logical, controls messaging. Default
TRUE.