Disperse: Processing Data.

In bioinfoDZ/RISC: Robust Integration of Single-Cell RNA-Seq Datasets

Processing Data.

Description

After data scaling, RISC will identify highly variably expressed genes, based on Quasi-Poinsson model, where the coefficient of variation is calculated for each gene (C > 0.5 as a cutoff for highly variable genes). Then, to controlling for the relationship between S (standard deviation) and mean (average value), Quasi-Poisson regression is used to further filter the genes with over- dispersion C caused by small mean. Lastly, RISC estimates the corresponding ratio with r between the observed C and the predicted C of each gene, with a threshold r > 1.

Usage

scDisperse(
  object,
  method = "loess",
  min.UMI = NULL,
  mean.cut = NULL,
  QP_bin = 100,
  lspan = 0.05,
  top.var = NULL,
  pval = 0.5
)

Arguments

object	RISC object: a framework dataset.
method	What method is used to define dispersion, now support "QP" and "loess", the default method is "loess".
min.UMI	A cutoff of the minimum UMIs of each gene, the genes expression below the min.UMI will be discarded from highly variable genes. The default value is 100 when input NULL.
mean.cut	A cutoff of the average value of each gene, the genes expression outside the mean.cut range will be discarded from highly variable genes. The input is a range vector, like c(0.1, 5).
QP_bin	The number of fragments using to fit dispersion in the Quasi-Poinsson model, how many bins are formed in the regression.
lspan	The number of parameter using to fit dispersion, controlling the degree of smoothing in the loess model.
top.var	The maximum number of highly variable genes, the default is NULL, including all the highly variable genes.
pval	The P-value is used to cut off the highly variable genes, the default is 0.5.

Value

RISC single cell dataset, the metadata slot.

References

Liu et al., Nature Biotech. (2021)

Examples

# RISC object
obj0 = raw.mat[[5]]
obj0 = scFilter(obj0, min.UMI = 0, max.UMI = Inf, min.gene = 10, min.cell = 3)
obj0 = scNormalize(obj0)
obj0 = scDisperse(obj0)

bioinfoDZ/RISC documentation built on March 30, 2024, 9:19 p.m.