performance.pcor: Quality of estimated partial correlations

Description

This function computed various performance measures of the estimated matrix of partial correlations.

Usage

performance.pcor(inferred.pcor, true.pcor=NULL,
			fdr=TRUE, cutoff.ggm=0.8,verbose=FALSE,plot.it=FALSE)

Arguments

inferred.pcor

matrix of estimated partial correlations

true.pcor

true matrix of partial correlations. Default is true.pcor=NULL

fdr

logical. If fdr=TRUE, the entries of inferred.pcor are tested for significance. Default is fdr=TRUE

cutoff.ggm

default cutoff for significant partial correlations. Default is cutoff.ggm=0.8

verbose

Print information on test results etc.. Default is verbose=FALSE

plot.it

Plot test results and ROC-curves. Default is plot.it=FALSE

Value

num.selected: number of selected edges
adj: binary matrix that encodes the existence of an edge between two nodes.
connectivity: vector of length ncol(inferred.pcor). Its ith entry indicated the number of nodes that are connected to the ith node.
positive.cor: percentage of positive partial correlations out of all selected edges.
power: power (if true.pcor is provided)
ppv: positive predictive value (if true.pcor is provided)
tpr: true positive rate (=power) (if true.pcor is provided)
fpr: true positive rate (=power) (if true.pcor is provided)
auc: area under the curve (if true.pcor is provided and fdr=TRUE)
TPR: vector of true positive rates corresponding to varying cut-offs (if true.pcor is provided and fdr=TRUE)
FPR: vector of false positive rates corresponding to varying cut-offs (if true.pcor is provided and fdr=TRUE)

Details

This function computes a range of performance measures: The function always returns the number of selected edges, the binary matrix that encodes the edges, the connectivity and the percentage of positive correlations. If true.pcor is provided, the function also returns the power (= true positive rate), the false positive rate and the positive predictive value. For non-sparse estimates that involve testing (i.e. fdr=TRUE) the function also returns the area under the curve, and a pair of vectors of false and true positive rates. The latter can e.g. be used to plot a ROC-curve.

References

N. Kraemer, J. Schaefer, A.-L. Boulesteix (2009) "Regularized Estimation of Large-Scale Gene Regulatory Networks using Gaussian Graphical Models", BMC Bioinformatics, 10:384

http://www.biomedcentral.com/1471-2105/10/384/