Skip to contents

Compute Standard r^2 LD Matrix

Source: R/compute_ld.R
compute_r2.Rd

Fast C++/Armadillo implementation of the standard pairwise squared Pearson correlation (r^2) for a window of SNP columns. Missing genotypes (NA) are mean-imputed per column before computation. This is the default LD metric in LDxBlocks and is 10-50x faster than stats::cor().

Usage

compute_r2(X, digits = -1L, n_threads = 1L)

Arguments

X

NumericMatrix (individuals x SNPs). Values 0/1/2. NA allowed.

digits

Integer. Rounding decimal places. -1 (default) = none.

n_threads

Integer. OpenMP threads. Default 1.

Value

Symmetric p x p NumericMatrix, diagonal 0, values in [0, 1].

When to use r^2 vs rV^2

r^2

Use for large unstructured datasets (> 500 k markers), random mating populations, or whenever speed matters. The standard estimator is inflated in related populations (i.e. will over-estimate LD) but this usually leads to slightly more conservative (larger) blocks rather than catastrophically wrong ones.

rV^2

Use for highly structured / related populations (livestock, inbred lines, family-based human cohorts) where kinship inflation would meaningfully distort block boundaries. Requires computing and inverting the GRM - prohibitive beyond ~5 k individuals.

See also

compute_rV2, compute_ld

Examples

set.seed(1)
G <- matrix(sample(0:2, 80 * 30, replace = TRUE), 80, 30)
r2 <- compute_r2(G, digits = 6L)
range(r2)   # [0, 1]
#> [1] 0.000000 0.117788