GGtools (version 5.8.0)

eqtlTests: compute association statistics between all probes and SNP in an smlSet instance

Description

compute association statistics (or point estimates and standard errors) between all probes and SNP in an smlSet instance, using out-of-memory storage; the basic test statistics are generated by the snp.rhs.tests function of the snpStats package

Usage

eqtlTests(smlSet, rhs = ~1 - 1, runname = "foo", 
	targdir = "foo", geneApply = lapply, 
	shortfac = 100, 
	checkValid = TRUE, useUncertain = TRUE, 
	glmfamily = "gaussian", doFFSUMM = FALSE)
eqtlEstimates(smlSet, rhs = ~1 - 1, runname = "foo", 
	targdir = "fooe", geneApply = lapply, 
	shortfac = 10000, 
	checkValid = TRUE, useUncertain = TRUE, 
	glmfamily = "gaussian")

Arguments

smlSet

instance of smlSet

rhs

fragment of a standard formula, minus a dependent variable (i.e., starts with tilde); bindings will be sought in pData(smlSet)

runname

string used to identify output ff files

targdir

string naming the folder where ff outputs will reside

geneApply

analog to lapply to drive iteration over probes

shortfac

ff contents will be multiplied by this quantity and stored as short integers

checkValid

logical, will apply validObject to smlSet if TRUE

useUncertain

logical, passed as uncertain parameter to snp.rhs.tests to specify whether uncertain genotypes will be used (as 'dosage' in GLM fitting)

glmfamily

family specification for snp.rhs.tests

doFFSUMM

logical indicating whether ff archives will be retained for col.summary outputs for SNPs

Value

returns an instance of eqtlTestsManager

Details

The purpose of the eqtlTests function is to allow very substantial eQTL search processes to occur with R. For several million SNP and tens of thousands of probes, the storage of test results requires attention to parsimony. The storage occurs out of memory, using the ff package, and employs short integers to represent chi squared statistics. These are scaled up prior to storage, and will be scaled down prior to use.

eqtlEstimates will use compact storage for both the point estimates and standard errors of association estimated under an additive genetic model

Note: snp.rhs.estimates will emit a warning whenever the parameters are not estimable. These warning are suppressed by temporary setting of options()[['warn']] to zero.

Examples

hm2ceuSMS = getSS("GGtools", c("20"), renameChrs=c("chr20"))
library(illuminaHumanv1.db)
cptag = get("CPNE1", revmap(illuminaHumanv1SYMBOL))
indc = which(featureNames(hm2ceuSMS) == cptag[1])
#
# get a set of additional genes on chr20
all20 = get("20", revmap(illuminaHumanv1CHR))
g20 = unique(c(all20[1:10], cptag))
#
hm = hm2ceuSMS[probeId(g20),]  # reduce problem
td = tempdir()
curd = getwd()
setwd(td)
time.lapply = unix.time(e1 <- eqtlTests( hm, ~male ))
time.lapply
e1
# best chisq(1) for CPNE1
topFeats(probeId(cptag), e1)
setwd(curd)