PrecTpFpMatrix {qtlhot}

Determine false positive and true positive rates for known targets.

Description

Determine how well different tests do to predict candidates of regulation.

Usage

FitAllTests(cross, pheno1, pheno2, Q.chr, Q.pos, verbose = TRUE)
JoinTestOutputs(comap, tests, file)
PrecTpFpMatrix(alpha, val.targets, all.orfs, tests, cand.reg, cis.cand.reg)
p.adjust.np(tests, method = "BH")

Arguments

cross

object of class cross

pheno1

first phenotype column number or character string name

pheno2

second phenotype column number or character string name; if more than one, then all phenotypes will be tested against pheno1

Q.chr

QTL chromosome (number or label)

Q.pos

QTL position in cM

verbose

verbose printout if TRUE

comap

list result of GetComappingTraits

alpha

significance levels at which summaries are computed

val.targets

validated targets of candidate regulators

all.orfs

all trait names

tests

list object as list of FitAllTests results, or of joined output created by JoinTestsOutputs

file

prefix for file names when running FitAllTests in parallel and saving test results in separate files

cand.reg

object from GetCandReg

cis.cand.reg

object from GetCisCandReg

method

method for p-value adjustment; see p.adjust

Details

FitAllTests invokes 7 tests. The hidden routine CitTests is invoked by call to FitAllTests; this is hidden because we do not recommend its use.

JoinTestOutputs joins results of FitAllTests, either from a list tests or from a collection of files prefixed by file. The joined tests from JoinTestOutputs are summarized with PrecTpFpMatrix using the biologically validated true positives, false positives and precision, for the inferred causal relations. We define a true positive as a statistically significant causal relation between a gene and a putative target gene when the putative target gene belongs to the known signature of the gene. Similarly, we define a false positive as a statistically significant causal relation between a gene and a putative target gene when the target gene does not belong to the signature. (For the AIC and BIC methods that do not provide a p-value measuring the significance of the causal call, we simply use the detected causal relations in the computation of true and false positives). The validated precision is computed as the ratio of true positives by the sum of true and false positives. The PrecTpFpMatrix computes these measures to both all genes, and to cis genes only. Simulations suggest only non-parametric tests need to be adjusted using Benjamini-Hochberg via p.adjust.np.

Value

List containing

Prec1, Prec2

matrix of precision with rows for significance level and columns for test; first is for all, second is for cis candidates only

Tp1, Tp2

matrix of true positive rate with rows for significance level and columns for test; first is for all, second is for cis candidates only

Fp1, Fp2

matrix of false positive rate with rows for significance level and columns for test; first is for all, second is for cis candidates only

Author(s)

Elias Chaibub Neto

See Also

GetCandReg, CMSTtests, p.adjust

Examples

example(GetCandReg)
## Suppose y1 is causal with targets y2 and y3.
targets <- list(y1 = c("y2","y3"))

tests <- list()
for(k in seq(names(comap.targets))) {
  tests[[k]] <- FitAllTests(CMSTCross, pheno1 = names(comap.targets)[k],
                      pheno2 = comap.targets[[k]],
                      Q.chr = cand.reg[k, 4],
                      Q.pos = cand.reg[k, 5])
}
names(tests) <- names(comap.targets)
tests <- JoinTestOutputs(comap.targets, tests)

PrecTpFpMatrix(alpha = seq(0.01, 0.10, by = 0.01),
  val.targets = targets, all.orfs = CMSThigh$names, tests = tests,
  cand.reg = cand.reg, cis.cand.reg = cis.cand.reg)