Skip to contents

Clique-Based LD Block Detection

Source: R/CLQD.R
CLQD.Rd

Partitions SNPs in a genomic window into LD cliques. Three algorithms are available via CLQmode:

"Density" (default)

Standard Big-LD clique scoring – edges built from the full r? matrix, cliques found by Bron-Kerbosch (igraph::max_cliques), scored by density = size / span. Exact, but exponential worst-case. Use checkLargest = TRUE to guard against blowup on dense WGS panels.

"Maximal"

Same algorithm as Density but scored by clique size only (original Big-LD maximality criterion).

"Louvain"

Louvain community detection (igraph::cluster_louvain). Polynomial time O(n log n) – never exponential. Recommended for WGS panels (>500 SNPs per window) or any time CLQmode = "Density" is slow. Communities are returned as bin IDs matching the Density output format so all downstream code is unaffected.

"Leiden"

Leiden community detection (igraph::cluster_leiden). Stricter than Louvain – produces well-connected communities without disconnected nodes. Requires igraph >= 1.3.0. Slightly slower than Louvain but higher quality communities. Recommended over Louvain when resolution matters.

When max_bp_distance is supplied (> 0), only SNP pairs within that physical distance have their r? computed (compute_r2_sparse_cpp). This reduces the O(p?) LD computation to near-O(p) for WGS panels where long-range LD is negligible (typically > 500 kb). Applied before any clique algorithm.

Usage

CLQD(
  subgeno,
  subSNPinfo,
  adj_subgeno,
  CLQcut = 0.5,
  clstgap = 40000,
  CLQmode = c("Density", "Maximal", "Louvain", "Leiden"),
  codechange = FALSE,
  checkLargest = FALSE,
  split = FALSE,
  digits = -1L,
  n_threads = 1L,
  max_bp_distance = 0L,
  verbose = FALSE
)

Arguments

subgeno

Numeric matrix (individuals x SNPs), 0/1/2 coded.

subSNPinfo

Data frame: col 1 = rsID, col 2 = bp position.

adj_subgeno

n x p matrix from prepare_geno().

CLQcut

Numeric in (0,1]. LD threshold for edges. Default 0.5.

clstgap

Integer. Max bp gap within clique (split=TRUE).

CLQmode

Algorithm: "Density" (default, exact, exponential worst-case), "Maximal" (exact, maximality criterion), "Louvain" (polynomial, recommended for WGS), "Leiden" (polynomial, stricter communities, igraph >= 1.3.0).

codechange

Logical. Sign-flip heuristic. Default FALSE.

checkLargest

Logical. Dense-core pre-pass for >= 500 SNPs when using Density or Maximal mode. Default FALSE. Has no effect when CLQmode is Louvain or Leiden (already polynomial).

split

Logical. Split cliques over gaps. Default FALSE.

digits

Integer. Rounding for LD. -1 = none (default).

n_threads

Integer. OpenMP threads. Default 1.

max_bp_distance

Integer. Maximum base-pair distance between a SNP pair for its r? to be computed. Pairs beyond this distance are assumed to be in negligible LD and set to 0. Default 0L = disabled (compute all pairs, original behaviour). Recommended value for WGS panels: 500000L (500 kb). Has no effect when the window spans less than max_bp_distance.

verbose

Logical.

Value

Integer vector length p: clique/community ID per SNP, NA = singleton.