combine.moments2: Builds a matrix of second moments for a set of individuals,...
In gtx: Genetics ToolboX

Description Usage Arguments Details Value Author(s) Examples

Given matrices of non-central second moments between a set of variables, calculated for two or more subsets of individuals, the matrix of non-central second moments between that set of variables, in the combined set of individuals, is obtained simply by addition. This function provides a convenience interface that deals with different row and column orderings in the input matrices, and also allows a subset of variables to be treated as pertaining to specific subsets of individuals only.

1	combine.moments2(xtxlist, fixed)

`xtxlist`	a named list of matrices of second moments.
`fixed`	names of variables for which regression coefficients are considered fixed across all studies.

The names of the elements of xtxlist are used as identifiers for the subsets of individuals for which each matrix has been calculated.

Variables whose names are included in the fixed argument must be included in all matrices in xtxlist, and are assumed to have identical definitions in all individuals. The intended application is that a regression model will be fitted with a coefficient for each of these variables that takes a single fixed value for all individuals.

Any variables included in matrices in xtxlist whose names are not included in the fixed argument are assumed to have potentially different definitions in the subsets of individuals corresponding to the matrices in xtxlist. An example would be ancestry principal component covariates, that might have the same names (“PC1”, “PC2”, ...) but do not have the same definitions in different subsets of individuals. The intended application is that a regression model will be fitted with a subset-specific coefficient for each of these variables, and the variables are therefore renamed by prepending identifiers constucted using the names of the elements of xtxlist.

The combined matrix has no “ONE” intercept term, and instead has length(xtxlist) subset-specific intercepts (named by prepending identifiers constucted using the names of the elements of xtxlist). This allows the the full rank of the combined matrix to be maintained while treating all subsets symetrically.

A numeric matrix of second moments.

Toby Johnson Toby.x.Johnson@gsk.com

data(mthfrex)
## artifical example with two datasets obtained by
## splitting mthfrex by the HTN variable

xtx.hi <- make.moments2(mthfr.params, c("SBP", "SexC", "Age"),
                        as.snpdata(list(snpinfo = mthfrex$snpinfo,
                                   data = subset(mthfrex$data, HTN == 1))))

xtx.lo <- make.moments2(mthfr.params, c("SBP", "SexC", "Age"),
                        as.snpdata(list(snpinfo = mthfrex$snpinfo,
                                   data = subset(mthfrex$data, HTN == 0))))

## make list of X'X matrices
xtx.list <- list(hi = xtx.hi, lo = xtx.lo)

## combine for outcome SBP and fixed effects for all SNPs
## other variables SexC and Age will be treated as study-specific
fixed <- paste(mthfr.params$snp, mthfr.params$coded.allele, sep = "_")
xtx.comb <- combine.moments2(xtx.list, c("SBP", fixed))

## fit regression model
n.comb <- sum(diag(xtx.comb)[1:length(xtx.list)])
lm.moments2(xtx.comb, "SBP", c("hi_ONE", "lo_ONE", "rs4846052_T"), n = n.comb)

## equivalent regression directly using subject-specific data
coeff.extract(lm(SBP ~ HTN + rs4846052_C, data = mthfrex$data))