Estimate SNP heritability using LDscore regression for a single annotation

An R implementation of the LD score regression method to estimate SNP heritability, mimicking ldsc --h2 from the ldsc package.

LDscores for the European subset of 1000g (2015 release) for 1,290,028 HapMap3 SNPs (with 1,173,569 SNPs with freq > 5%) are bundled within the ldsR package and are used by default. This corresponds to the eur_w_ld_chr folder previously shared at the LDSC github.

You can inspect the LDscores used by default: arrow::read_parquet(system.file("extdata", "eur_w_ld.parquet", package = "ldsR"))

ldsc_h2 does not perform any quality control on the input summary statistics, except to merge with the 1,290,028 HapMap3 SNPs in the reference panel. See the munge() function to mimic the munge_sumstats.py function.

Usage

ldsc_h2(
  sumstat,
  pop_prev = NULL,
  sample_prev = 0.5,
  weights = NULL,
  M = NULL,
  n_blocks = 200
)

Arguments

sumstat

A dplyr::tibble() with columns SNP, Z and N

pop_prev

prevalence of the disorder in the general population

sample_prev

the prevalence of the disorder in the sample. Default value is 0.5, reflecting a case-control study using effective N as sample size

weights

Optional, a data.frame or tbl with columns SNP, L2

M

Optional, the number of SNPs in the reference panel

n_blocks

Number of blocks to use for the jackknife estimator

Value

a dplyr::tibble() with columns h2 and h2_se

Examples

p <- system.file("extdata", "eur_w_ld.parquet", package = "ldsR")
snps <- arrow::read_parquet(p, col_select = c("SNP"))
snps$N <- 130000
snps$Z <- rnorm(nrow(snps))
ldsc_h2(snps)
#> # A tibble: 1 × 6
#>          h2   h2_se   int  int_se mean_chi2 lambda_gc
#>       <dbl>   <dbl> <dbl>   <dbl>     <dbl>     <dbl>
#> 1 -0.000113 0.00110  1.00 0.00312     0.999     0.996