hrunbiasedTesting: Cox proportional hazard model for gene signatures testing

Description Usage Arguments Details Value Author(s) References See Also Examples

View source: R/hrunbiasedTesting.r

Description

Cox model for a target signature of genes allowing several correction types and testing procedures

Usage

1
2
3
4
5
6
7
hrunbiasedTesting(nda, gene.signature, alternative ="two.sided",
           test.type = c("asymptotic", "random.signatures", "random.genes"),
           correction.type = c("NCsignatures", "Gsignature", "RUV", "none"),
           score = NULL, seqquant = 0.25,  nrs = 5000,
           comp.RUV = c(1,2), adj.var.GS = "GSadj",
           adjusted.var.random = NA, adjusted.var.fixed = NA, mc.cores = 1,
           signature.method = "zscore", maxGenes = 5000, executation.info=TRUE, ...)

Arguments

nda

ExpressionSet object. pData(nda) must contain survival information in attributes evn (numeric vector with event description - 0 no event 1 event) and tev (time to event)

gene.signature

Vector with the names of the gene signature to be tested (must match with rownames(nda))

alternative

Character string specifying the alternative hypothesis, must be one of "two.sided" (default), "greater" or "less"

test.type

Character string specifying the reference distribution to be used, must include at least one of "asymptotic" or "random.signatures" (all three by default)

correction.type

Standard Cox model can be used when correction.type = "none", otherwise correction for global signature "Gsignature", correction for negative control signatures "NCsignatures" or factor analysis "RUV" are available

score

Vector of the same length as number of rows in nda with scores to be used for "NCsignatures" and "RUV" corrections

seqquant

Value between zero and one used to set negative controls

nrs

Number of random signatures to be used when test.type = "random.signatures"

comp.RUV

Factors used in factor analysis signature correction when correction.type = "RUV"

adj.var.GS

Only if correction.type is not "none" and siganture.method is "zscore". Transforming signature scores by subtracting the correction variable scores "GScor", residuals of linear model using the correction variable as covariate "GSlmcor", or using the correction variable in the Cox model "GSadj" (default)

adjusted.var.random

Variables that enter in the model as random effect

adjusted.var.fixed

Confounding variables that enter in the model as fixed effect

mc.cores

The number of cores to use for parallel executions

signature.method

Signature score method, to choose from "zscore", "gsva" or "plage"

maxGenes

Number of genes to be used to determine reference distribution if test.type = "random.genes"

executation.info

If TRUE information regarding executation stages is shown

...

Arguments passed to or from other methods to the low level.

Details

Testing statistical significance of the relationship between a gene signature and recurrence can be done by considering both asymptotic distributions or random signature based distributions. The interpretation of the two is different: asymptotic testing finds evidence against the null hypothesis HR = 1; random signatures quantifies how special is the relation of a gene signature to recurrence in comparison to any possible random set of genes.

A useful correction we consider is the adjusment by a global signature which uses the information of all genes and that may point towards technical / sampling directions. This global signature can be influenced by both technical and biological effects. In order to reduce the biological impact, adjusting variables are determined by negative control signatures based on a score ordered from low to high. For example, we suggest to use a measure of dispersion so that only low-variable genes are used in an adjusting signature. Other scores based on cross-study reproducibility or any sort of biological prior information could also be employed. See hrunbiasedDiagnostic for more details in selecting the correction type.

For seqquant = 0, then the correction type coincides with correction.type = "none" and for seqquant = 1 then the correction type coincides with correction.type = "Gsignature".

Value

An object of class hrunbiasedTesting containing the following items

"hr.ts"

Cox model outcome of target signature

"hr.rs"

Cox model outcome of random signatures

"hr.tg"

Cox model outcomes of genes composing the target signature

"hr.rg"

Cox model outcomes of random genes

"pvals.rs"

P-value found using as reference distribution the statistics obtained for random signatures

"test.type"

Reference distribution used for hypothesis testing

"gene.signature"

Gene signature used for hypothesis testing

Author(s)

Adria Caballe Mestres

References

Caballe Mestres A, Berenguer Llergo A and Stephan-Otto Attolini C. Adjusting for systematic technical biases in risk assessment of gene signatures in transcriptomic cancer cohorts. bioRxiv (2018).

Gagnon-Bartsch J. and Speed T. Using control genes to correct for unwanted variation in microarray data. Biostatistics 13, 539-552 (2012).

Hanzelmann S., Castelo R. and Guinney J. GSVA: gene set variation analysis for microarray and RNA-Seq data. BMC Bioinformatics 7 (2013).

See Also

plot.hrunbiasedTesting, randomSigCoxmeSimple and hrunbiasedDiagnostic

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
eh <- ExperimentHub()
nda.brca <- query(eh, "mcsurvdata")[["EH1497"]]
nda.gse1456 <- nda.brca[,nda.brca$dataset=="GSE1456"]
data(mammaprint.sig)
mammaprint.sig <- intersect(mammaprint.sig, rownames(nda.gse1456))
correction.type <-  "Gsignature"
tout.GS <- hrunbiasedTesting(nda.gse1456,
    gene.signature = mammaprint.sig,
    test.type = c("asymptotic", "random.signatures", "random.genes"),
    correction.type = "Gsignature", mc.cores=1,
    nrs = 100,  executation.info = FALSE)
tout.GS

adricaba/hrunbiased documentation built on May 24, 2019, 7:48 a.m.