corMeta.GLS.invert.mats: The efficient meta-analysis estimator accounting for...

In tamartsi/MetaCor: Performs Stratified SNP-trait analysis and combines the results using the MetaCor method

Description

Computes the fixed effects meta-analysis estimator while accounting for correlations between effect estimates in the variance estimation. The computation is done for many SNPs at a time using matrix operations, including inversion of many matrices at once.

Usage

corMeta.GLS.invert.mats(beta, var, cov)

Arguments

beta	a p by S matrix of estimated effects of each of p SNPs on a trait in each of S strata.
var	a p by S matrix of variances of the estimated effects of each of p SNPs on a trait in each of S strata.
cov	a p by (S*(S-1)/2) matrix representing p vectors of covariances between the estimated effects of each of p SNPs on a trait in each of S strata.

Details

Computes the MetaGLS (MetaCor1 form the MetaCor paper) estimator and test. This function can be memory intensive since it inverts many matrices together, in recurssion. Don't use for more than 6 strata.

Value

A data frame with p rows, with beta, var, test.stat and pval providing the meta-analysis effect estimates, their variances, test statistics, and p-values for each of the SNPs. beta, cov and var should have appropriate column names.

Note

This function is called by MetaCor and it is advisable to use MetaCor. It can also be called directly, but then the test of heterogeneity is not performed. Also, MetaCor allows for selection of strata while this function does not.

Author(s)

Tamar Sofer

Examples

### create a small data set with the right formating:
require(mvtnorm)
n.strat <- 5
n.est <- 5
## create a positive definite matrix to be a covariance matrix
sigma <- matrix(rnorm(n.strat^2), n.strat, n.strat)
sigma <- t(sigma) 

effects <- rmvnorm(n.est, c(2,2,3,3,0), sigma)
colnames(effects) <- paste0("Beta.s", 1:n.strat)
vars <- matrix(rep(diag(sigma), n.est), nrow = n.est, ncol = n.strat, byrow = T)
colnames(vars) <- paste0("var.s", 1:n.strat)
covs <- matrix(rep(sigma[lower.tri(sigma)], n.est), nrow = n.est, byrow = T)
colnames(covs) <- c(paste0("cov.s1:", paste0("s", 2:n.strat)), paste0("cov.s2:", paste0("s", 3:n.strat)), paste0("cov.s3:", paste0("s", 4:n.strat)) , "cov.s4:s5")

## test:
corMeta.GLS.invert.mats(effects, vars, covs) 

tamartsi/MetaCor documentation built on May 31, 2019, 2:56 a.m.