permutationPval: calculate permutation pval for association between X and Z if...

In vitkl/NetFeaturePval: Find Features Enriched among Nodes that Interact with Each Single Node

Description

calculate permutation pval for association between X and Z if both are connected to Y

Usage

permutationPval(interactions2permute = nodeX ~ nodeY,
  associations2test = nodeX ~ nodeZ, node_attr = NULL, data,
  statistic = nodeX + nodeZ ~ .N, select_nodes = NULL, N = 1000,
  cores = NULL, cluster_type = "PSOCK", seed = NULL,
  also_permuteYZ = F, clustermq = F, clustermq_mem = 4000,
  clustermq_jobs = 100, clustermq_template = list(),
  split_comp_inner_N = NULL, clustermq_fail_on_error = TRUE,
  clustermq_log_worker = FALSE)

Arguments

interactions2permute	formula specifying columns that contain X-Y interactions that will be permuted
associations2test	formula specifying columns that contain X-Z correspondence: empirical p-values for the association of X with Z will be tested, NOTE: Y-Z interactions are assumed
node_attr	formula or list of formulas specifying columns that contain attributes of X, Y or Z or their combination (X ~ degree, X + Z ~ pvalue)
data	data.table containing interaction data and attributes
statistic	formula that specifies how to calculate statistic using attibutes from node_attr by node X in interactions2permute and Z in associations2test, details: permutationPvalHelper. In data.table synthax: DT[, (observed/permuted)statistic := eval(right-hand-side of formula), by = .(eval(column names in the left-hand-side of formula))]
select_nodes	formula or list of formulas specifying which nodes of specific node type to select before permutation based on condition (X ~ degree > 10)
N	number of times to run permutation of PPI network
cores	specify how many cores to use for parallel processing, default (NULL) is to detect all cores on the machine and use all minus one. When using LSF cluster you must specify the number of cores to use because detectCores doen't know how much cores you have requested from LSF (with bsub -n) and detects all cores on the actual physical node.
cluster_type	Type of the cluster to create when using R parrallel (clusterApply). Type "FORK" means cluster nodes share objects in memory. Details: (makeCluster)
seed	seed for RNG for reproducible sampling
also_permuteYZ	logical, permute Y-Z interactions in addition to X-Y (specified in interactions2permute) ?
clustermq	if TRUE uses clustermq job scheduling (Q) instead of local parallelisation (parReplicate) = F, = 4000, = 100, clustermq_template = list(), split_comp_inner_N
clustermq_mem	memory in MB to allocate for each job (ignored unless clustermq == TRUE)
clustermq_jobs	maximal number of computing cluster jobs to use (ignored unless clustermq == TRUE)
clustermq_template	Add specific arguments to computing cluster job submission call. Not needed in most cases. Details: Q (ignored unless clustermq == TRUE)
split_comp_inner_N	parallel evaluation of permutations is split into the outer and inner replicate calls helps to save memory by decreasing the total size of the result. This argument let's you manually specify the number of inner replicate calls. This has to be optimised for data size when using clustermq (ignored unless clustermq == TRUE)
clustermq_fail_on_error	If TRUE clustermq will fail if one of the jobs returns an error. Details: Q (ignored unless clustermq == TRUE)
clustermq_log_worker	If TRUE clustermq will save log of worker jobs. Where in is save is determined by clustermq.template
formula	argument for permutationPvalPlot, formula specifying attribute of which nodes to plot like this: nodeX + nodeZ ~ p.value. The default is to plot p.value histogram for nodeX and nodeZ as specified in the x object
x	argument for permutationPvalPlot, output of permutationPval, class "XYZinteration_XZEmpiricalPval"
...	argument for permutationPvalPlot, base R plotting parameters

Value

object of S3 class "XYZinteration_XZEmpiricalPval" (list), containing permutationPval function call, standardised node names, and data.table containing the original data but appended with empirical p-value (p.value), observed_statistic, YmissingZ_perX, and higher_counts, not_missing used to calculate p-value

Author(s)

Vitalii Kleshchevnikov

Examples

## Not run: 
res = permutationPval(interactions2permute = nodeX ~ nodeY,
 associations2test = nodeX ~ nodeZ, node_attr = NULL,
 data, statistic = nodeX + nodeZ ~ .N, select_nodes = NULL, N = 1000,
 cores = NULL, seed = NULL, also_permuteYZ = F, clustermq = F)

# print
res

# plot p-value distribution (hist)
plot(res)

# plot the number of Y without Z per X (hist),
# formula is used to subset the table before plotting
# to avoid plotting single number multiple times
plot(res, nodeX ~ YmissingZ_perX)

## End(Not run)

vitkl/NetFeaturePval documentation built on May 19, 2019, 9:39 p.m.