CORREP (version 1.38.0)

cor.LRtest.std: Maximum Likehood Ratio Test for Multivariate Correlation Estimator

Description

Performs LRT to test if multivariate correlation vanishes. Note this code `standardizes' estimated correlation matrix to make sure its determinant is positive.

Usage

cor.LRtest.std(x, m1, m2)

Arguments

x

data matrix, column represents samples (conditions), and row represents variables (genes), see example below for format information

m1

number of replicates for gene X

m2

number of replicates for gene Y

Value

p.value

The p-value of the LRT

Details

Under the multivaraite normal distribution assumption, the column vector of the data is iid sample. We test the followiing hypothesis: H0: Z ~ N(mu, Sigma0), H1 ~ N(mu, Sigma1). Let M = Inverse(Sigma0)*Sigma1, the likelihood ratio test statistic $G^2$, is, n[trace(M)-log(det(M))-(m1+m2)]. Under the H0, $G^2$ follows a chi-square distribution with $2m1*m2$ degree of freedom. In some case, the determinant of M is negative so that the log(det(M)) return NaN. There are two ways to deal with this problem, one, those M's whose determinant are negative tend to consist of very small correlations that are biological irrelevant. Therefore, we can simply ignore those gene pairs that the determinant correlation matrix M is negative. Second, we can `standardize' the M to make the determinant positive (implemented in function cor.LRtest.std). In most of cases, we recommend using function cor.LRtest. Use cor.LRtest.std only you know what you are doing.

References

Zhu, D and Li Y. 2007. Multivariate Correlation Estimator for Inferring Functional Relationships from Replicated 'OMICS' data. Submitted.

See Also

cor.LRtest, cor.test

Examples

library("CORREP")
library("MASS")
Sigma <- matrix(c(1, 0.8, .5,.5, 0.8, 1,
0.5,0.5,0.5,0.5,1,0.6,0.5,0.5,0.6,1),4,4)
dat <- mvrnorm(50, mu=c(0,0,0,0), Sigma)
dat.std <- apply(dat,2,function(x) x/sd(x))
cor.LRtest.std(t(dat.std), m1=2, m2=2)