twilight: Estimation of the local false discovery rate

In twilight: Estimation of local false discovery rate

Description

The function performs the successive exclusion procedure (SEP) as described in Scheid and Spang (2004).

Usage

twilight(xin, lambda = NULL, B = 0, boot.ci = 0.95, clus = NULL, verbose = TRUE)

Arguments

xin	Numerical vector of p-values or a twilight object.
lambda	Numerical value denoting the penalty factor. If not specified, the function searchs for an appropriate regularization parameter.
B	Numerical value specifying the number of bootstrap samples. If not specified, no bootstrap estimates are calculated.
boot.ci	Numerical value denoting the probability value for bootstrap confidence intervals of local false discovery rate and prior pi0.
clus	A list or numerical value to be passed to makeCluster(clus,...) in library(snow). If specified, bootstrapping is performed in parallel. No checks on clus are performed. Please make sure that makeCluster(clus) works properly in your environment.
verbose	Logical value for message printing.

Details

In short, the successive exclusion procedure divides the set of p-values into two parts. The first part is chosen such that it resembles a uniform distribution while containing as many p-values as possible. This set of p-values represents the set of p-values derived from non-induced genes. The height of the uninform distribution is a natural estimate for the mixture parameter pi0. The p-value not contained in the uniform part serve as representatives of p-values derived from induced genes. Their distribution is the basis of the final estimate of the local false discovery rate.

NOTE: Library(snow) has to be loaded manually. It is not loaded as part of 'suggests' or 'depends' because it is only available under UNIX. If twilight does not work with the current version of snow, please send a report.

Value

Returns a twilight object consisting of a data.frame named result with variables

pvalue	Sorted input vector.
qvalue	q-values computed as described in Storey and Tibshirani (2003) with new estimate pi0.
index	Index of the original ordering.
fdr	Local false discovery rate averaged over 10 runs of SEP.
mean.fdr	Bootstrap estimate of local false discovery rate.
lower.fdr	Lower boot.ci-bootstrap confidence bound.
upper.fdr	Upper boot.ci-bootstrap confidence bound.

Values are sorted by pvalue.

Note

Additional output consists of

lambda	Regularization parameter.
pi0	SEP estimate of prior probability.
boot.pi0	Bootstrap estimate and boot.ci-bootstrap confidence bounds.
boot.ci	Passes boot.ci for plotting.
effect	Histogram of effect size distributions averaged over 10 runs of SEP.

If xin is of class twilight, the remaining slots are filled with corresponding input values. If xin is not of class twilight, these slots remain empty.

Author(s)

Stefanie Scheid

References

Scheid S and Spang R (2004): A stochastic downhill search algorithm for estimating the local false discovery rate, IEEE TCBB 1(3), 98–108.

Scheid S and Spang R (2005): twilight; a Bioconductor package for estimating the local false discovery rate, Bioinformatics 21(12), 2921–2922.

Scheid S and Spang R (2006): Permutation filtering: A novel concept for significance analysis of large-scale genomic data, in: Apostolico A, Guerra C, Istrail S, Pevzner P, and Waterman M (Eds.): Research in Computational Molecular Biology: 10th Annual International Conference, Proceedings of RECOMB 2006, Venice, Italy, April 2-5, 2006. Lecture Notes in Computer Science vol. 3909, Springer, Heidelberg, pp. 338-347.

Storey JD and Tibshirani R (2003): Statistical significance for genomewide studies, PNAS 100(16), 9440–9445.

See Also

twilight.pval, plot.twilight, exfdr

Examples

### twilight object created with B=1000 bootstrap samples
data(exfdr) 
print(exfdr)
plot(exfdr)

twilight documentation built on Nov. 8, 2020, 5:38 p.m.