A GenotypeMixture uses several genotype panels (see
readGenotypes) as one LD reference whose LD is the
weighted average of theirs, \(\sum_k w_k R_k\). That is the LD of
summary statistics meta-analysed across cohorts, with each cohort's
weight its share of the GWAS sample size.
Unlike an LdMatrix mixture, which is combined once when it
is built, the panels stay lazy and on disk: LD is read from each and
combined only for the variants a step asks about, and every panel keeps
answering for its own samples. A step that needs allele frequencies,
missingness or minor-allele counts judges each panel on its own and
drops a variant that fails in any of them.
It is a MultiAssayExperiment with one experiment per
cohort: that cohort's genotype panel, cut to the variants every panel
carries, in the first panel's order. The mixture's variants, as
variantIds gives them, are the first panel's. Each panel
keeps its own samples and allele coding, and its rowData column
mixtureSign is 1 where it codes a variant's alleles as the
first panel does and -1 where it codes them the other way round.
The sampleMap names each sample <cohort>_<sampleId>, since
cohorts can reuse sample ids, and the weights are
metadata(x)$mixtureWeights (see mixtureWeights).
Row subsetting, x[i, ] by position, variant id or GRanges,
cuts every panel to the same variants, and column subsetting narrows
the panels' samples. A panel cannot be dropped, since the weights
describe all of them: build a new GenotypeMixture over the panels
wanted instead.
See also
GenotypeMixture to build one, LdMatrix.