MetaCor: Combine stratum-specific effect estimates in meta-analysis...

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

Description

Given association analysis results and covariances between the stratum-specific estimates, perform meta-analysis using MetaCor.

Usage

MetaCor(dat, block.size = 50000, testType = "MetaGLS", subgroup.vec = NULL, test.heterogeneity = TRUE, verbose = FALSE)

Arguments

dat	A matrix with rows corresponding to m SNPs, and columns with the following information (note that this matrix is prepared by the function stratLMMTest that calls this function) ID for each SNP, columns with names "Beta.strat_name", "var.strat_name", "cov.strat_name:other_strat_name" corresponding to estimated SNP effect, variances of these estimators, and covariances between them, for a few strata (at least 2).
block.size	Number of SNPs to meta-analyze together (to speed computation). Used only if MetaGLS is testType. Default is 50,000.
testType	Either MetaGLS (MetaCor2 from paper) or MetaCor.fixef (MetaCor1 from paper). If an unfamiliar test name is specified, will be taken to be MetaGLS.
subgroup.vec	If one wants to meta-analyze only some of the strata, specify the names in this vector of strings.
test.heterogeneity	Should a test of heterogeneity between stratum-specific effects be performed? default is yes.
verbose	If verbose, print progress updates.

Value

return a matrix with information for each SNP - meta.beta, meta.se, meta.test and meta.pval are the meta-analyzed effect sizes, standard errors, test statistics and p-values. Similar information is given for the test of heterogeneity using hetero.pval, etc.

Author(s)

Tamar Sofer

Examples

 ### This function is used by stratLMMTest when combining the effect estimates from all strata. Example of usage outside of stratLMMTest:
  
### 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")
dat <- cbind(data.frame(snpID = 1:n.est), effects, vars, covs)

## test:
MetaCor(dat) ## uses MetaGLS, MetaCor2 from the paper
MetaCor(dat, testType = "MetaCor.fixef") ## MetaCor1 from paper
MetaCor(dat, testType = "MetaCor.fixed") ## if wrong name of test, it uses MetaGLS

## can choose to meta-analyze only part of the strata:
MetaCor(dat, subgroup.vec = c("s1", "s2", "s5"), testType = "MetaCor.fixef")

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