mvTest-method: Multivariate tests on results from 'dream()'

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

Multivariate tests on results from dream()

Description

Evaluate multivariate tests on results from dream() using vcov() to compute the covariance between estimated regression coefficients across multiple responses. A joint test to see if the coefficients are jointly different from zero is performed using meta-analysis methods that account for the covariance.

Usage

mvTest(
  fit,
  vobj,
  features,
  coef,
  method = c("FE.empirical", "FE", "RE2C", "tstat", "hotelling", "sidak", "fisher"),
  shrink.cov = TRUE,
  BPPARAM = SerialParam(),
  ...
)

## S4 method for signature 'MArrayLM,EList,vector'
mvTest(
  fit,
  vobj,
  features,
  coef,
  method = c("FE.empirical", "FE", "RE2C", "tstat", "hotelling", "sidak", "fisher"),
  shrink.cov = TRUE,
  BPPARAM = SerialParam(),
  ...
)

## S4 method for signature 'MArrayLM,EList,missing'
mvTest(
  fit,
  vobj,
  features,
  coef,
  method = c("FE.empirical", "FE", "RE2C", "tstat", "hotelling", "sidak", "fisher"),
  shrink.cov = TRUE,
  BPPARAM = SerialParam(),
  ...
)

## S4 method for signature 'MArrayLM,EList,list'
mvTest(
  fit,
  vobj,
  features,
  coef,
  method = c("FE.empirical", "FE", "RE2C", "tstat", "hotelling", "sidak", "fisher"),
  shrink.cov = TRUE,
  BPPARAM = SerialParam(),
  ...
)

## S4 method for signature 'mvTest_input,ANY,ANY'
mvTest(
  fit,
  vobj,
  features,
  coef,
  method = c("FE.empirical", "FE", "RE2C", "tstat", "hotelling", "sidak", "fisher"),
  shrink.cov = TRUE,
  BPPARAM = SerialParam(),
  ...
)

## S4 method for signature 'MArrayLM,matrix,ANY'
mvTest(
  fit,
  vobj,
  features,
  coef,
  method = c("FE.empirical", "FE", "RE2C", "tstat", "hotelling", "sidak", "fisher"),
  shrink.cov = TRUE,
  BPPARAM = SerialParam(),
  ...
)

Arguments

fit	MArrayLM or MArrayLM2 returned by dream()
vobj	matrix or EList object returned by voom()
features	a) indeces or names of features to perform multivariate test on, b) list of indeces or names. If missing, perform joint test on all features.
coef	name of coefficient or contrast to be tested
method	statistical method used to perform multivariate test. See details. 'FE' is a fixed effect test that models the covariance between coefficients. 'FE.empirical' use compute empirical p-values by sampling from the null distribution and fitting with a gamma. 'RE2C' is a random effect test of heterogeneity of the estimated coefficients that models the covariance between coefficients, and also incorporates a fixed effects test too. 'tstat' combines the t-statistics and models the covariance between coefficients. 'hotelling' performs the Hotelling T2 test. 'sidak' returns the smallest p-value and accounting for the number of tests. 'fisher' combines the p-value using Fisher's method assuming independent tests.
shrink.cov	shrink the covariance matrix between coefficients using the Schafer-Strimmer method
BPPARAM	parameters for parallel evaluation
...	other arugments

Details

See package remaCor for details about the remaCor::RE2C() test, and see remaCor::LS() for details about the fixed effect test. When only 1 feature is selected, the original p-value is returned and the test statistic is set to NA.

For the "RE2C" test, the final test statistic is the sum of a test statistic for the mean effect (stat.FE) and heterogeneity across effects (stat.het). mvTest() returns 0 if stat.het is negative in extremely rare cases.

Value

Returns a data.frame with the statistics from each test, the pvalue from the test, n_features, method, and lambda from the Schafer-Strimmer method to shrink the estimated covariance. When shrink.cov=FALSE, lambda = 0.

Examples

# library(variancePartition)
library(edgeR)
library(BiocParallel)

data(varPartDEdata)

# normalize RNA-seq counts
dge <- DGEList(counts = countMatrix)
dge <- calcNormFactors(dge)

# specify formula with random effect for Individual
form <- ~ Disease + (1 | Individual)

# compute observation weights
vobj <- voomWithDreamWeights(dge[1:20, ], form, metadata)

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

# Multivariate test of features 1 and 2
mvTest(fit, vobj, 1:2, coef = "Disease1")

# Test multiple sets of features
lst <- list(a = 1:2, b = 3:4)
mvTest(fit, vobj, lst, coef = "Disease1", BPPARAM = SnowParam(2))

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