calculateQValue: Compute saps q-value

Description

This function computes the saps q-value for a candidate prognostic geneset by computing the saps score for randomly generated genesets and determining the proportion at least as significant as the saps score for the candidate set. This function is normally called by saps.

Usage

calculateQValue(dataSet, sampleSize, survivalTimes, followup, saps_score,
  random.samples, qvalue.samples, cpus, gsea.perm, rankedGenes)

Arguments

dataSet

A matrix, where the column names are gene identifiers and the values are gene expression levels. Each row should contain data for a single patient.

sampleSize

The desired size for the randomly sampled gene sets.

survivalTimes

A vector of survival times. The length must equal the number of rows in dataSet.

followup

A vector of 0 or 1 values, indicating whether the patient was lost to followup (0) or not (1). The length must equal the number of rows (i.e. patients) in dataSet.

saps_score

The saps score for the candidate geneset. The q-value is calculated as the proportion of random saps scores whose absolute value is >= to the provided saps score.

random.samples

The number of random gene sets to sample during the calculation of P_random.

qvalue.samples

The number of random gene sets to sample for purposes of computing the q-value.

cpus

An integer that specifies the number of cpus/cores to be used when calculating P_enrichment. If greater than 1, the snowfall package must be installed or an error will occur.

gsea.perm

The number of permutations to be used when calculating P_enrich. This is passed to the runGSA function in the piano package.

rankedGenes

A vector of ranking scores for each gene in dataSet. Ordinarily this will be the z-scores obtained by a call to rankConcordance.

Value

The function returns a list with two elements:
q_valuethe calculated q-value.
random_saps_scoresa vector of individual saps scores for each randomly generated geneset.

References

Beck AH, Knoblauch NW, Hefti MM, Kaplan J, Schnitt SJ, et al. (2013) Significance Analysis of Prognostic Signatures. PLoS Comput Biol 9(1): e1002875.doi:10.1371/journal.pcbi.1002875

Examples

Run this code

# 25 patients, none lost to followup
followup <- rep(1, 25)

# first 5 patients have good survival (in days)
time <- c(25, 27, 24, 21, 26, sample(1:3, 20, TRUE))*365

# create data for 100 genes, 25 patients
dat <- matrix(rnorm(25*100), nrow=25, ncol=100)
colnames(dat) <- as.character(1:100)

# borderline significant saps score
saps_score <- 1.3

rankedGenes <- rankConcordance(dat, time, followup)[,"z"]

q_value <- calculateQValue(dat, 5, time, followup, saps_score, random.samples=100,
     qvalue.samples=10, cpus=1, gsea.perm=1000, rankedGenes)

q_value$q_value
random_scores <- abs(q_value$random_saps_scores)
hist(random_scores)
length(random_scores[random_scores > saps_score])