Estimate the Single-Variant Inflation Factor from a Scan

Given a single-variant association scan, estimate its genomic-control inflation factor by one of two methods, after the standard LDSC QC (common variants; the MHC excluded). The returned factor is the input to the GLOW calibration methods: calibrate_pvalues (M1) and the z_scale = 1/sqrt(factor) hook in glow_region_test (M2).

estimate_inflation_factor(
  scan,
  method = c("ldsc_intercept", "lambda_gc"),
  ld_scores = NULL,
  pvalue_col = "pvalue",
  z_col = "Z",
  maf_col = "MAF",
  maf_min = 0.01,
  exclude_mhc = TRUE,
  mhc_chr = "6",
  mhc_start = 2.5e+07,
  mhc_end = 3.5e+07,
  winsorize = 80
)

Arguments

scan

A single-variant scan data.frame (the marginal_scan shape). "ldsc_intercept" needs columns chr, pos, ref, alt and z_col; "lambda_gc" needs pvalue_col. maf_col is used for QC when present.

method

"ldsc_intercept" (default) or "lambda_gc".

ld_scores

LD-score table from compute_ld_scores (chr, pos, ref, alt, ld); required for "ldsc_intercept", ignored for "lambda_gc".

pvalue_col, z_col

Column names for the p-value ("lambda_gc") and the Z-score ("ldsc_intercept", \(\chi^2 = Z^2\)). Defaults "pvalue", "Z".

maf_col, maf_min

Minor-allele-frequency column and threshold for the common-variant QC (default "MAF", 0.01). QC is skipped if maf_col is absent or maf_min <= 0.

exclude_mhc, mhc_chr, mhc_start, mhc_end

Exclude the MHC region (default TRUE; chr "6", 25e6-35e6, hg38).

winsorize

\(\chi^2\) cap passed to ldsc_regression (default 80).

Value

A one-row data.frame: method, factor, n_variants, and intercept, intercept_se, slope, slope_se, confounding_ratio, ratio_se, lambda_gc, mean_chi2 (the LDSC fields are NA for "lambda_gc"). The factor column is the calibration factor; the row doubles as a provenance record to persist alongside a run.

Details

"ldsc_intercept" (the principled, confounding-only factor)

The scan is merged to ld_scores by chr:pos:ref:alt, QC'd (\(\mathrm{MAF} \ge\)maf_min; the MHC dropped; positive finite LD scores), and ldsc_regression is run on \(\chi^2 = Z^2\) vs the LD score; the intercept is the factor.

"lambda_gc" (removes all inflation)

The genomic-control \(\lambda_{\mathrm{GC}}\) via compute_lambda_gc on the scan's p-values (after the same MAF QC). No LD scores needed.

Applying the same MAF QC to both methods keeps them comparable as common-variant inflation estimates; set maf_min = 0 (or omit maf_col) to use all variants.

See also

ldsc_regression, compute_ld_scores, compute_lambda_gc, calibrate_pvalues

Examples

if (FALSE) { # \dontrun{
scan <- read.csv("marginal_all.csv")
ld   <- readRDS("ldscores.rds")
est  <- estimate_inflation_factor(scan, "ldsc_intercept", ld_scores = ld)
est$factor                       # the LDSC intercept 
estimate_inflation_factor(scan, "lambda_gc")$factor
} # }