consensusKME: Calculate consensus kME (eigengene-based connectivities) across multiple data sets.

Description

Calculate consensus kME (eigengene-based connectivities) across multiple data sets, typically following a consensus module analysis.

Usage

consensusKME(
  multiExpr,
  moduleLabels, 
  multiEigengenes = NULL, 
  consensusQuantile = 0, 
  signed = TRUE,
  useModules = NULL,
  metaAnalysisWeights = NULL,
  corAndPvalueFnc = corAndPvalue, corOptions = list(), corComponent = "cor",
  getQvalues = FALSE,
  useRankPvalue = TRUE,
  rankPvalueOptions = list(calculateQvalue = getQvalues, pValueMethod = "scale"),
  setNames = NULL, 
  excludeGrey = TRUE, greyLabel = ifelse(is.numeric(moduleLabels), 0, "grey"))

Arguments

multiExpr

Expression (or other numeric) data in a multi-set format. A vector of lists; in each list there must be a component named `data' whose content is a matrix or dataframe or array of dimension 2.

moduleLabels

Module labels: one label for each gene in multiExpr.

multiEigengenes

Optional eigengenes of modules specified in moduleLabels. If not given, will be calculated from multiExpr.

signed

logical: should the network be considered signed? In signed networks (TRUE), negative kME values are not considered significant and the corresponding p-values will be one-sided. In unsigned networks (FALSE), negative kME

useModules

Optional specification of module labels to which the analysis should be restricted. This could be useful if there are many modules, most of which are not interesting. Note that the "grey" module cannot be used with useModules.

consensusQuantile

Quantile for the consensus calculation. Should be a number between 0 (minimum) and 1.

metaAnalysisWeights

Optional specification of meta-analysis weights for each input set. If given, must be a numeric vector of length equal the number of input data sets (i.e., length(multiExpr)). These weights will be used in addition to constant weights and wei

corAndPvalueFnc

Function that calculates associations between expression profiles and eigengenes. See details.

corOptions

List giving additional arguments to function corAndPvalueFnc. See details.

corComponent

Name of the component of output of corAndPvalueFnc that contains the actual correlation.

getQvalues

logical: should q-values (estimates of FDR) be calculated?

useRankPvalue

Logical: should the rankPvalue function be used to obtain alternative meta-analysis statistics?

rankPvalueOptions

Additional options for function rankPvalue. These include na.last (default "keep"), ties.method (default "average"), calculateQvalue (defa

setNames

names for the input sets. If not given, will be taken from names(multiExpr). If those are NULL as well, the names will be "Set_1", "Set_2", ....

excludeGrey

logical: should the grey module be excluded from the kME tables? Since the grey module is typically not a real module, it makes little sense to report kME values for it.

greyLabel

label that labels the grey module.

Value

Data frame with the following components (for easier readability the order here is not the same as in the actual output):
IDGene ID, taken from the column names of the first input data set
consensus.kME.1, consensus.kME.2, ...Consensus kME (that is, the requested quantile of the kMEs in the individual data sets)in each module for each gene across the input data sets. The module labels (here 1, 2, etc.) correspond to those in moduleLabels.
weightedAverage.equalWeights.kME1, weightedAverage.equalWeights.kME2, ...Average kME in each module for each gene across the input data sets.
weightedAverage.RootDoFWeights.kME1, weightedAverage.RootDoFWeights.kME2, ...Weighted average kME in each module for each gene across the input data sets. The weight of each data set is proportional to the square root of the number of samples in the set.
weightedAverage.DoFWeights.kME1, weightedAverage.DoFWeights.kME2, ...Weighted average kME in each module for each gene across the input data sets. The weight of each data set is proportional to number of samples in the set.
weightedAverage.userWeights.kME1, weightedAverage.userWeights.kME2, ...(Only present if input metaAnalysisWeights is non-NULL.) Weighted average kME in each module for each gene across the input data sets. The weight of each data set is given in metaAnalysisWeights.
meta.Z.equalWeights.kME1, meta.Z.equalWeights.kME2, ...Meta-analysis Z statistic for kME in each module, obtained by weighing the Z scores in each set equally. Only returned if the function corAndPvalueFnc returns the Z statistics corresponding to the correlations.
meta.Z.RootDoFWeights.kME1, meta.Z.RootDoFWeights.kME2, ...Meta-analysis Z statistic for kME in each module, obtained by weighing the Z scores in each set by the square root of the number of samples. Only returned if the function corAndPvalueFnc returns the Z statistics corresponding to the correlations.
meta.Z.DoFWeights.kME1, meta.Z.DoFWeights.kME2, ...Meta-analysis Z statistic for kME in each module, obtained by weighing the Z scores in each set by the number of samples. Only returned if the function corAndPvalueFnc returns the Z statistics corresponding to the correlations.
meta.Z.userWeights.kME1, meta.Z.userWeights.kME2, ...Meta-analysis Z statistic for kME in each module, obtained by weighing the Z scores in each set by metaAnalysisWeights. Only returned if metaAnalysisWeights is non-NULL and the function corAndPvalueFnc returns the Z statistics corresponding to the correlations.
meta.p.equalWeights.kME1, meta.p.equalWeights.kME2, ...p-values obtained from the equal-weight meta-analysis Z statistics. Only returned if the function corAndPvalueFnc returns the Z statistics corresponding to the correlations.
meta.p.RootDoFWeights.kME1, meta.p.RootDoFWeights.kME2, ...p-values obtained from the meta-analysis Z statistics with weights proportional to the square root of the number of samples. Only returned if the function corAndPvalueFnc returns the Z statistics corresponding to the correlations.
meta.p.DoFWeights.kME1, meta.p.DoFWeights.kME2, ...p-values obtained from the degree-of-freedom weight meta-analysis Z statistics. Only returned if the function corAndPvalueFnc returns the Z statistics corresponding to the correlations.
meta.p.userWeights.kME1, meta.p.userWeights.kME2, ...p-values obtained from the user-supplied weight meta-analysis Z statistics. Only returned if metaAnalysisWeights is non-NULL and the function corAndPvalueFnc returns the Z statistics corresponding to the correlations.
meta.q.equalWeights.kME1, meta.q.equalWeights.kME2, ...q-values obtained from the equal-weight meta-analysis p-values. Only present if getQvalues is TRUE and the function corAndPvalueFnc returns the Z statistics corresponding to the kME values.
meta.q.RootDoFWeights.kME1, meta.q.RootDoFWeights.kME2, ...q-values obtained from the meta-analysis p-values with weights proportional to the square root of the number of samples. Only present if getQvalues is TRUE and the function corAndPvalueFnc returns the Z statistics corresponding to the kME values.
meta.q.DoFWeights.kME1, meta.q.DoFWeights.kME2, ...q-values obtained from the degree-of-freedom weight meta-analysis p-values. Only present if getQvalues is TRUE and the function corAndPvalueFnc returns the Z statistics corresponding to the kME values.
meta.q.userWeights.kME1, meta.q.userWeights.kME2, ...q-values obtained from the user-specified weight meta-analysis p-values. Only present if metaAnalysisWeights is non-NULL, getQvalues is TRUE and the function corAndPvalueFnc returns the Z statistics corresponding to the kME values.
The next set of columns contain the results of function rankPvalue and are only present if input useRankPvalue is TRUE. Some columns may be missing depending on the options specified in rankPvalueOptions. We explicitly list columns that are based on weighing each set equally; names of these columns carry the suffix .equalWeights
pValueExtremeRank.ME1.equalWeights, pValueExtremeRank.ME2.equalWeights, ...This is the minimum between pValueLowRank and pValueHighRank, i.e. min(pValueLow, pValueHigh)
pValueLowRank.ME1.equalWeights, pValueLowRank.ME2.equalWeights, ...Asymptotic p-value for observing a consistently low value across the columns of datS based on the rank method.
pValueHighRank.ME1.equalWeights, pValueHighRank.ME2.equalWeights, ...Asymptotic p-value for observing a consistently low value across the columns of datS based on the rank method.
pValueExtremeScale.ME1.equalWeights, pValueExtremeScale.ME2.equalWeights, ...This is the minimum between pValueLowScale and pValueHighScale, i.e. min(pValueLow, pValueHigh)
pValueLowScale.ME1.equalWeights, pValueLowScale.ME2.equalWeights, ...Asymptotic p-value for observing a consistently low value across the columns of datS based on the Scale method.
pValueHighScale.ME1.equalWeights, pValueHighScale.ME2.equalWeights, ...Asymptotic p-value for observing a consistently low value across the columns of datS based on the Scale method.
qValueExtremeRank.ME1.equalWeights, qValueExtremeRank.ME2.equalWeights, ...local false discovery rate (q-value) corresponding to the p-value pValueExtremeRank
qValueLowRank.ME1.equalWeights, qValueLowRank.ME2.equalWeights, ...local false discovery rate (q-value) corresponding to the p-value pValueLowRank
qValueHighRank.ME1.equalWeights, lueHighRank.ME2.equalWeights, ...local false discovery rate (q-value) corresponding to the p-value pValueHighRank
qValueExtremeScale.ME1.equalWeights, qValueExtremeScale.ME2.equalWeights, ...local false discovery rate (q-value) corresponding to the p-value pValueExtremeScale
qValueLowScale.ME1.equalWeights, qValueLowScale.ME2.equalWeights, ...local false discovery rate (q-value) corresponding to the p-value pValueLowScale
qValueHighScale.ME1.equalWeights,qValueHighScale.ME2.equalWeights, ...local false discovery rate (q-value) corresponding to the p-value pValueHighScale
...Analogous columns corresponding to weighing individual sets by the square root of the number of samples, by number of samples, and by user weights (if given). The corresponding column name suffixes are .RootDoFWeights, .DoFWeights, and .userWeights.
The following set of columns summarize kME in individual input data sets.
kME1.Set_1, kME1.Set_2, ..., kME2.Set_1, kME2.Set_2, ...kME values for each gene in each module in each given data set.
p.kME1.Set_1, p.kME1.Set_2, ..., p.kME2.Set_1, p.kME2.Set_2, ...p-values corresponding to kME values for each gene in each module in each given data set.
q.kME1.Set_1, q.kME1.Set_2, ..., q.kME2.Set_1, q.kME2.Set_2, ...q-values corresponding to kME values for each gene in each module in each given data set. Only returned if getQvalues is TRUE.
Z.kME1.Set_1, Z.kME1.Set_2, ..., Z.kME2.Set_1, Z.kME2.Set_2, ...Z statistics corresponding to kME values for each gene in each module in each given data set. Only present if the function corAndPvalueFnc returns the Z statistics corresponding to the kME values.

Details

The function corAndPvalueFnc is currently is expected to accept arguments x (gene expression profiles), y (eigengene expression profiles), and alternative with possibilities at least "greater", "two.sided". Any additional arguments can be passed via corOptions.

The function corAndPvalueFnc should return a list which at the least contains (1) a matrix of associations of genes and eigengenes (this component should have the name given by corComponent), and (2) a matrix of the corresponding p-values, named "p" or "p.value". Other components are optional but for full functionality should include (3) nObs giving the number of observations for each association (which is the number of samples less number of missing data - this can in principle vary from association to association), and (4) Z giving a Z static for each observation. If these are missing, nObs is calculated in the main function, and calculations using the Z statistic are skipped.

References

Langfelder P, Horvath S., WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics. 2008 Dec 29; 9:559.