LD clumping by a per-variant score using bigsnpr — ld_clump_by

Wraps bigsnpr::snp_clumping with the boilerplate of wrapping a numeric dosage matrix into a bigstatsr::FBM.code256 object and of handling the common pitfall of a single-variant input.

Usage

ld_clump_by_score(
  X,
  score,
  chr,
  pos,
  r2 = 0.2,
  window_kb = 100/r2,
  verbose = FALSE
)

Arguments

X: Numeric matrix of 0/1/2 allele dosages, n rows by p variants. Column names are expected to be variant IDs but are not required.
score: Numeric vector of length ncol(X). Higher values favour retention during clumping (e.g. -log10 p, |Z|, MAF). May be NULL, in which case bigsnpr falls back to minor allele frequency computed from X.
chr: Integer or character vector of length ncol(X) giving the chromosome for each variant.
pos: Integer vector of length ncol(X) giving the base-pair position for each variant.
r2: Numeric in (0, 1]. r-squared threshold for clumping (variants within window_kb whose r2 exceeds r2 and have lower score are removed). Default 0.2.
window_kb: Numeric. Window size in kilobases. Default is 100 / r2, matching the common "ld-clump size = 100/r2" heuristic used in many GWAS pipelines.
verbose: Logical. If TRUE, print the number of retained variants. Default FALSE.

Value

An integer vector of indices (into X columns) kept after clumping. For a single-column X, returns 1L.

Examples

if (FALSE) { # \dontrun{
  set.seed(1)
  n <- 500; p <- 20
  X <- matrix(rbinom(n * p, 2, 0.3), n, p)
  colnames(X) <- paste0("chr1:", seq_len(p) * 1000, ":A:G")
  s <- runif(p)
  chr <- rep(1L, p); pos <- seq_len(p) * 1000L
  keep <- ld_clump_by_score(X, score = s, chr = chr, pos = pos, r2 = 0.2)
} # }