diffVar-method: Test differential variance

In GabrielHoffman/variancePartition: Quantify and interpret drivers of variation in multilevel gene expression experiments

Test differential variance

Description

Test the association between a covariate of interest and the response's deviation from expectation.

Usage

diffVar(
  fit,
  method = c("AD", "SQ"),
  scale = c("leverage", "none"),
  BPPARAM = SerialParam(),
  ...
)

## S4 method for signature 'MArrayLM'
diffVar(
  fit,
  method = c("AD", "SQ"),
  scale = c("leverage", "none"),
  BPPARAM = SerialParam(),
  ...
)

Arguments

fit	model fit from dream()
method	transform the residuals using absolute deviation ("AD") or squared deviation ("SQ").
scale	scale each observation by "leverage", or no scaling ("none")
BPPARAM	parameters for parallel evaluation
...	other parameters passed to dream()

Details

This method performs a test of differential variance between two subsets of the data, in a way that generalizes to multiple categories, continuous variables and metrics of spread beyond variance. For the two category test, this method is simular to Levene's test. This model was adapted from Phipson, et al (2014), extended to linear mixed models, and adapted to be compatible with dream().

This method is composed of multiple steps where 1) a typical linear (mixed) model is fit with dream(), 2) residuals are computed and transformed based on an absolute value or squaring transform, 3) a second regression is performed with dream() to test if a variable is associated with increased deviation from expectation. Both regression take advantage of the dream() linear (mixed) modelling framework followed by empirical Bayes shrinkage that extends the limma::voom() framework.

Note that diffVar() takes the results of the first regression as a parameter to use as a starting point.

Value

MArrayLM object storing differential results to be passed to topTable()

References

\insertRef

phipson2014diffvarvariancePartition

See Also

missMethyl::diffVar(), car::leveneTest()

Examples

# library(variancePartition)
library(edgeR)
data(varPartDEdata)

# filter genes by number of counts
isexpr <- rowSums(cpm(countMatrix) > 0.1) >= 5

# Standard usage of limma/voom
dge <- DGEList(countMatrix[isexpr, ])
dge <- calcNormFactors(dge)

# make this vignette faster by analyzing a subset of genes
dge <- dge[1:1000, ]

# regression formula
form <- ~Disease

# estimate precision weights
vobj <- voomWithDreamWeights(dge, form, metadata)

# fit dream model
fit <- dream(vobj, form, metadata)
fit <- eBayes(fit)

# fit differential variance model
res <- diffVar(fit)

# extract results for differential variance based on Disease
topTable(res, coef = "Disease1", number = 3)

# Box plot of top hit
# Since ASCL3 has a negative logFC,
# the deviation from expectation is *smaller* in
# Disease==1 compared to baseline.
gene <- "ENST00000325884.1 gene=ASCL3"
boxplot(vobj$E[gene, ] ~ metadata$Disease, main = gene)

GabrielHoffman/variancePartition documentation built on April 20, 2024, 7:29 p.m.