Calculation of a consensus network (topological overlap).
consensusTOM(
# Supply either ...
# ... information needed to calculate individual TOMs
multiExpr, # Data checking options
checkMissingData = TRUE,
# Blocking options
blocks = NULL,
maxBlockSize = 5000,
blockSizePenaltyPower = 5,
nPreclusteringCenters = NULL,
randomSeed = 54321,
# Network construction arguments: correlation options
corType = "pearson",
maxPOutliers = 1,
quickCor = 0,
pearsonFallback = "individual",
cosineCorrelation = FALSE,
replaceMissingAdjacencies = FALSE,
# Adjacency function options
power = 6,
networkType = "unsigned",
checkPower = TRUE,
# Topological overlap options
TOMType = "unsigned",
TOMDenom = "min",
suppressNegativeTOM = FALSE,
# Save individual TOMs?
saveIndividualTOMs = TRUE,
individualTOMFileNames = "individualTOM-Set%s-Block%b.RData",
# ... or individual TOM information
individualTOMInfo = NULL,
useIndivTOMSubset = NULL,
##### Consensus calculation options
useBlocks = NULL,
networkCalibration = c("single quantile", "full quantile", "none"),
# Save calibrated TOMs?
saveCalibratedIndividualTOMs = FALSE,
calibratedIndividualTOMFilePattern = "calibratedIndividualTOM-Set%s-Block%b.RData",
# Simple quantile calibration options
calibrationQuantile = 0.95,
sampleForCalibration = TRUE, sampleForCalibrationFactor = 1000,
getNetworkCalibrationSamples = FALSE,
# Consensus definition
consensusQuantile = 0,
useMean = FALSE,
setWeights = NULL,
# Return options
saveConsensusTOMs = TRUE,
consensusTOMFilePattern = "consensusTOM-Block%b.RData",
returnTOMs = FALSE,
# Internal handling of TOMs
useDiskCache = NULL, chunkSize = NULL,
cacheDir = ".",
cacheBase = ".blockConsModsCache",
nThreads = 1,
# Diagnostic messages
verbose = 1,
indent = 0)
List with the following components:
only present if input returnTOMs is TRUE. A list containing consensus TOM
for each block, stored as a distance structure.
only present if input saveConsensusTOMs is TRUE. A vector of file names, one for
each block, in which the TOM for the corresponding block is stored. TOM is saved as a distance structure to
save space.
a copy of the input saveConsensusTOMs.
information about individual set TOMs. A copy of the input individualTOMInfo
if given; otherwise the result of calling blockwiseIndividualTOMs. See blockwiseIndividualTOMs for
details.
Further components are retained for debugging and/or convenience.
a copy of the input useIndivTOMSubset.
a list containing information about which samples and genes are "good" in the sense
that they do not contain more than a certain fraction of missing data and (for genes) have non-zero variance.
See goodSamplesGenesMS for details.
number of "good" genes in goodSamplesGenes above.
number of input sets.
a copy of the input saveCalibratedIndividualTOMs.
if input saveCalibratedIndividualTOMs is TRUE, this
component will contain the file names of calibrated individual networks. The file names are arranged in a
character matrix with each row corresponding to one input set and each column to one block.
if input getNetworkCalibrationSamples is TRUE, a list with one
component per block. Each component is in turn a list with two components: sampleIndex is a vector
contain the indices of the TOM samples (the indices refer to a flattened distance structure), and
TOMSamples is a matrix of TOM samples with each row corresponding to a sample in sampleIndex,
and each column to one input set.
a copy of the input consensusQuantile.
A vector of length nSets that
contains, for each set, the number of (calibrated) elements that were less than or equal the consensus for that element.
expression data in the multi-set format (see checkSets). A vector of
lists, one per set. Each set must contain a component data that contains the expression data, with
rows corresponding to samples and columns to genes or probes.
logical: should data be checked for excessive numbers of missing entries in genes and samples, and for genes with zero variance? See details.
optional specification of blocks in which hierarchical clustering and module detection
should be performed. If given, must be a numeric vector with one entry per gene
of multiExpr giving the number of the block to which the corresponding gene belongs.
integer giving maximum block size for module detection. Ignored if blocks
above is non-NULL. Otherwise, if the number of genes in datExpr exceeds maxBlockSize, genes
will be pre-clustered into blocks whose size should not exceed maxBlockSize.
number specifying how strongly blocks should be penalized for exceeding the
maximum size. Set to a lrge number or Inf if not exceeding maximum block size is very important.
number of centers for pre-clustering. Larger numbers typically results in better
but slower pre-clustering. The default is as.integer(min(nGenes/20, 100*nGenes/preferredSize))
and is an attempt to arrive at a reasonable number given the resources available.
integer to be used as seed for the random number generator before the function
starts. If a current seed exists, it is saved and restored upon exit. If NULL is given, the
function will not save and restore the seed.
character string specifying the correlation to be used. Allowed values are (unique
abbreviations of) "pearson" and "bicor", corresponding to Pearson and bidweight
midcorrelation, respectively. Missing values are handled using the pariwise.complete.obs option.
only used for corType=="bicor". Specifies the maximum percentile of data
that can be considered outliers on either
side of the median separately. For each side of the median, if
higher percentile than maxPOutliers is considered an outlier by the weight function based on
9*mad(x), the width of the weight function is increased such that the percentile of outliers on
that side of the median equals maxPOutliers. Using maxPOutliers=1 will effectively disable
all weight function broadening; using maxPOutliers=0 will give results that are quite similar (but
not equal to) Pearson correlation.
real number between 0 and 1 that controls the handling of missing data in the calculation of correlations. See details.
Specifies whether the bicor calculation, if used, should revert to Pearson when
median absolute deviation (mad) is zero. Recongnized values are (abbreviations of)
"none", "individual", "all". If set to
"none", zero mad will result in NA for the corresponding correlation.
If set to "individual", Pearson calculation will be used only for columns that have zero mad.
If set to "all", the presence of a single zero mad will cause the whole variable to be treated in
Pearson correlation manner (as if the corresponding robust option was set to FALSE). Has no
effect for Pearson correlation. See bicor.
logical: should the cosine version of the correlation calculation be used? The cosine calculation differs from the standard one in that it does not subtract the mean.
soft-thresholding power for network construction.
network type. Allowed values are (unique abbreviations of) "unsigned",
"signed", "signed hybrid". See adjacency.
logical: should basic sanity check be performed on the supplied power? If
you would like to experiment with unusual powers, set the argument to FALSE and proceed with
caution.
logical: should missing values in the calculation of adjacency be replaced by 0?
one of "none", "unsigned", "signed", "signed Nowick",
"unsigned 2", "signed 2" and "signed Nowick 2". If "none", adjacency
will be used for clustering. See TOMsimilarityFromExpr for details.
a character string specifying the TOM variant to be used. Recognized values are
"min" giving the standard TOM described in Zhang and Horvath (2005), and "mean" in which
the min function in the denominator is replaced by mean. The "mean" may produce
better results but at this time should be considered experimental.
Logical: should the result be set to zero when negative? Negative TOM values can occur when
TOMType is "signed Nowick".
logical: should individual TOMs be saved to disk for later use?
character string giving the file names to save individual TOMs into. The
following tags should be used to make the file names unique for each set and block: %s will be
replaced by the set number; %N will be replaced by the set name (taken from names(multiExpr))
if it exists, otherwise by set number; %b will be replaced by the block number. If the file names turn
out to be non-unique, an error will be generated.
Optional data for TOM matrices in individual data sets. This object is returned by
the function blockwiseIndividualTOMs. If not given, appropriate topological overlaps will be
calculated using the network contruction options below.
If individualTOMInfo is given, this argument allows to only select a subset
of the individual set networks contained in individualTOMInfo. It should be a numeric vector giving the
indices of the individual sets to be used. Note that this argument is NOT applied to multiExpr.
optional specification of blocks that should be used for the calcualtions. The default is to use all blocks.
network calibration method. One of "single quantile", "full quantile", "none" (or a unique abbreviation of one of them).
logical: should the calibrated individual TOMs be saved?
pattern of file names for saving calibrated individual TOMs.
if networkCalibration is "single quantile",
topological overlaps (or adjacencies if
TOMs are not computed) will be scaled such that their calibrationQuantile quantiles will agree.
if TRUE, calibration quantiles will be determined from a sample of network
similarities. Note that using all data can double the memory footprint of the function and the function
may fail.
determines the number of samples for calibration: the number is
1/calibrationQuantile * sampleForCalibrationFactor. Should be set well above 1 to ensure accuracy of the
sampled quantile.
logical: should the sampled values used for network calibration be returned?
quantile at which consensus is to be defined. See details.
logical: should the consensus be determined from a (possibly weighted) mean across the data sets rather than a quantile?
Optional vector (one component per input set) of weights to be used for weighted mean
consensus. Only used when useMean above is TRUE.
logical: should the consensus topological overlap matrices for each block be saved and returned?
character string containing the file namefiles containing the
consensus topological overlaps. The tag %b will be replaced by the block number. If the resulting file
names are non-unique (for example, because the user gives a file name without a %b tag), an error
will be generated.
These files are standard R data files and can be loaded using the load
function.
logical: should calculated consensus TOM(s) be returned?
should calculated network similarities in individual sets be temporarilly saved
to disk? Saving to disk is somewhat slower than keeping all data in memory, but for large blocks and/or
many sets the memory footprint may be too big. If not given (the default), the function will determine
the need of caching based on the size of the data. See chunkSize below for additional information.
network similarities are saved in smaller chunks of size chunkSize. If NULL,
an appropriate chunk size will be determined from an estimate of available memory. Note that if the chunk size
is greater than the memory required for storing intemediate results, disk cache use will automatically be
disabled.
character string containing the directory into which cache files should be written. The user should make sure that the filesystem has enough free space to hold the cache files which can get quite large.
character string containing the desired name for the cache files. The actual file
names will consists of cacheBase and a suffix to make the file names unique.
non-negative integer specifying the number of parallel threads to be used by certain parts of correlation calculations. This option only has an effect on systems on which a POSIX thread library is available (which currently includes Linux and Mac OSX, but excludes Windows). If zero, the number of online processors will be used if it can be determined dynamically, otherwise correlation calculations will use 2 threads.
integer level of verbosity. Zero means silent, higher values make the output progressively more and more verbose.
indentation for diagnostic messages. Zero means no indentation, each unit adds two spaces.
Peter Langfelder
The function starts by optionally filtering out samples that have too many missing entries and genes
that have either too many missing entries or zero variance in at least one set. Genes that are filtered
out are left unassigned by the module detection. Returned eigengenes will contain NA in entries
corresponding to filtered-out samples.
If blocks is not given and
the number of genes exceeds maxBlockSize, genes are pre-clustered into blocks using the function
consensusProjectiveKMeans; otherwise all genes are treated in a single block.
For each block of genes, the network is constructed and (if requested) topological overlap is calculated in each set. To minimize memory usage, calculated topological overlaps are optionally saved to disk in chunks until they are needed again for the calculation of the consensus network topological overlap.
Before calculation of the consensus Topological Overlap, individual TOMs are optionally calibrated. Calibration methods include single quantile scaling and full quantile normalization.
Single quantile
scaling raises individual TOM in sets 2,3,... to a power such that the quantiles given by
calibrationQuantile agree with the quantile in set 1. Since the high TOMs are usually the most
important
for module identification, the value of calibrationQuantile is close to (but not equal) 1. To speed up
quantile calculation, the quantiles can be determined on a randomly-chosen component subset of the TOM
matrices.
Full quantile normalization, implemented in normalize.quantiles, adjusts the
TOM matrices such that all quantiles equal each other (and equal to the quantiles of the component-wise
average of the individual TOM matrices).
Note that network calibration is performed separately in each block, i.e., the normalizing transformation may differ between blocks. This is necessary to avoid manipulating a full TOM in memory.
The consensus TOM is calculated as the component-wise consensusQuantile quantile of the individual
(set) TOMs; that is, for each gene pair (TOM entry), the consensusQuantile quantile across all input
sets. Alternatively, one can also use (weighted) component-wise mean across all imput data sets.
If requested, the consensus topological overlaps are saved to disk for later use.
WGCNA methodology has been described in
Bin Zhang and Steve Horvath (2005) "A General Framework for Weighted Gene Co-Expression Network Analysis", Statistical Applications in Genetics and Molecular Biology: Vol. 4: No. 1, Article 17 PMID: 16646834
The original reference for the WGCNA package is
Langfelder P, Horvath S (2008) WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics 2008, 9:559 PMID: 19114008
For consensus modules, see
Langfelder P, Horvath S (2007) "Eigengene networks for studying the relationships between co-expression modules", BMC Systems Biology 2007, 1:54
This function uses quantile normalization described, for example, in
Bolstad BM1, Irizarry RA, Astrand M, Speed TP (2003) "A comparison of normalization methods for high density oligonucleotide array data based on variance and bias", Bioinformatics. 2003 Jan 22;19(2):1
blockwiseIndividualTOMs for calculation of topological overlaps across multiple sets.