stratLMMTest: Test genetic variants using a stratified linear mixed model

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

Description

Takes a study data - outcome, covariants, genetic data, covariance matrices, and a list defining strata - and perfomrs a stratified trait-genotypes association analysis (and test).

Usage

stratLMMTest(Y, W, G, covMatList, IDsList, blockSize = 5000, metaCorBlockSize = 50000, testType = "MetaGLS", verbose = TRUE)

Arguments

Y	An n vector of outcomes/trait values for n perticipants
W	An n by k matrix of covariates values corresponding to n participants. Row names should be the participants IDs.
G	An n by p matrix of p genotype values for each of n paticipants.
covMatList	A named list of matrices that model the correaltion structure between the participants (e.g. kiship/GRM matrix).
IDsList	A named list of participants IDd in strata. IDs can overlap between strata.
blockSize	Association analysis of blockSize SNPs is computed together to speed up comoputation.
metaCorBlockSize	MetaGLS (MetaCor2 in the paper) test statistics are calculated jointly for metaCorBlockSize SNPs to speed up computation.
testType	MetaCor test type - MetaCor.fixef (MetaCor1) or MetaGLS (MetaCor2).
verbose	If verbose is TRUE, progress is reported via messages.

Value

A matrix providing for each SNP the estimated effect size, SEs, and p-value in each stratum, as well as the meta-analyzed estimated effect sizes, SEs and p-values, and the results from a test of heterogeneity of effects between strata.

Author(s)

Tamar Sofer

Examples

#### load the data - note: all data in the package is simulated and is very simple. 
### load a data frame with genotype data
data(G)
G[1:3,]
## note: sex_stratInd are indicators for selection of a maximum set of participants if we want the two sex strata independent.

## load a data frame with covariates info
data(W)
head(W)

## load the outcome
data(Y)
head(Y)

## load the simplified kinship matrix
data(K)
K[1:5,1:5] ## uncorrelated individuals since they are from different families
K[which(W$ID_fam == 1),which(W$ID_fam == 1)] ## these individuals are from the same family and have kinship coefficient 0.5.

## Prepare to fitting a stratified model with strata of sex - create a list of participants IDs in each of the strata:
## check:
all(W$ID == G$ID) #TRUE
IDs.list.gender <- list(male = W$ID[which(W$sex == 1)], female = W$ID[which(W$sex == 0)])
IDs.list.gender.stratInd <- list(male = W$ID[which(W$sex == 1 & W$sex_stratInd == 1)], female = W$ID[which(W$sex == 0 & W$sex_stratInd == 1)])

### fit a stratified model for males and females separately, and combine the results using MetaCor using a single function that does it all:
strat.model <- stratLMMTest(Y, as.matrix(W[,c("intercept", "bmi", "bmi2", "age")]), as.matrix(G[,grep("SNP", colnames(G))]), covMatList = list(kinship = K), IDsList = IDs.list.gender, verbose = T, testType = "MetaGLS")

### When according to the kinship matrix, the two strata are indpendent... (and also trying to use MetaCor1, names here "MetaCor.fixef") 
strat.model <- stratLMMTest(Y, as.matrix(W[,c("intercept", "bmi", "bmi2", "age")]), as.matrix(G[,grep("SNP", colnames(G))]), covMatList = list(kinship = K), IDsList = IDs.list.gender.stratInd, verbose = T, testType = "MetaCor.fixef")

## if the list of IDs provides only a single stratum, a pooled model is fit:
pooled.model   <- stratLMMTest(Y,  as.matrix(W[,c("intercept", "bmi", "bmi2", "age", "sex")]), as.matrix(G[,grep("SNP", colnames(G))]), covMatList = list(kinship = K), IDsList = list(pooled = rownames(W)), verbose = T, testType = "MetaCor.fixef")

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