Perform statistical test to check whether gene expression is constant across pseudotime within a lineage

This test assesses whether average gene expression is associated with pseudotime.

associationTest(models, ...)

# S4 method for SingleCellExperiment
associationTest(
  models,
  global = TRUE,
  lineages = FALSE,
  l2fc = 0,
  nPoints = 2 * tradeSeq::nknots(models),
  contrastType = "start"
)

# S4 method for list
associationTest(models, global = TRUE, lineages = FALSE, l2fc = 0)

Arguments

models	The fitted GAMs, typically the output from fitGAM.
...	parameters including:
global	If TRUE, test for all lineages simultaneously.
lineages	If TRUE, test for all lineages independently.
l2fc	The log2 fold change threshold to test against. Note, that this will affect both the global test and the pairwise comparisons.
nPoints	The number of points used per lineage to set up the contrast. Defaults to 2 times the number of knots. Note that not all points may end up being actually used in the inference; only linearly independent contrasts will be used.
contrastType	The contrast used to impose the log2 fold-change threshold. Defaults to "start". Three options are possible: - If "start", the starting point of each lineage is used to compare against all other points, and the fold-change threshold is applied on these contrasts. - If "end", the procedure is similar to "start", except that the reference point will now be the endpoint of each lineage rather than the starting point. - If "consecutive", then consecutive points along each lineage will be used as contrasts. This is the original way the associationTest was implemented and is kept for backwards compatibility. If a fold change threshold has been set, we recommend users to use either the "start" or "end" options.

Value

A matrix with the wald statistic, the degrees of freedom and the (unadjusted) p-value associated with each gene for all the tests performed. If the testing procedure was unsuccessful for a particular gene, NA values will be returned for that gene.

Details

If a log2 fold-change threshold has not been set, we use the QR decompositon through qr.solve to invert the variance-covariance matrix of the contrasts. If instead a log2 fold chalnge-threshold has been set, we invert that matrix using the Cholesky decomposition through chol2inv.

Examples

set.seed(8)
data(crv, package="tradeSeq")
data(countMatrix, package="tradeSeq")
sce <- fitGAM(counts = as.matrix(countMatrix),
                  sds = crv,
                  nknots = 5)
assocRes <- associationTest(sce)