Skip to contents

Allele harmonization for Mendelian Randomization

harmonizeAlleles.Rd

Ensures that the effect allele coding in the instrument table (from GWAS) matches the allele coding in the genotype data at each site. Handles allele flipping (when alleles are swapped) and detects strand-ambiguous SNPs (A/T and G/C pairs) that cannot be reliably harmonized.

When alleles need to be flipped, beta_ZX is multiplied by -1 and EAF is replaced by 1 - EAF. Palindromic SNPs with EAF near 0.5 (within eafPalindromicThreshold of 0.5) are dropped because strand cannot be reliably inferred from allele frequency.

Usage

harmonizeAlleles(
  instrumentTable,
  genotypeAlleles,
  eafPalindromicThreshold = 0.08,
  cohortAlleleFrequencies = NULL
)

Arguments

instrumentTable

Data frame with columns: snp_id, effect_allele, other_allele, beta_ZX, se_ZX, eaf. Typically the output of getMRInstruments.

genotypeAlleles

Data frame with columns: snp_id, allele_coded, allele_noncoded. Describes the allele coding used in the genotype data at a particular site.

eafPalindromicThreshold

Numeric threshold for removing palindromic SNPs. Palindromic SNPs with EAF between (0.5 - threshold) and (0.5 + threshold) are removed. Default is 0.08 (i.e., EAF between 0.42 and 0.58).

cohortAlleleFrequencies

Optional named numeric vector of cohort allele frequencies, keyed by SNP ID. The frequency should be for the coded allele in the genotype data (i.e., mean(genotype)/2). Used to resolve palindromic SNPs when EAF is far enough from 0.5.

Value

A list with two elements:

instrumentTable

The harmonized instrument table with updated effect_allele, other_allele, beta_ZX, and eaf where allele flips were applied. A logical column flipped is added.

removedSnps

Data frame of SNPs that were removed, with a column reason indicating why (e.g., "palindromic_ambiguous_eaf", "allele_mismatch").

Details

Harmonize Alleles Between Instrument Table and Genotype Data

The harmonization algorithm proceeds as follows for each SNP:

  1. If the instrument effect_allele matches the genotype allele_coded: no action needed.

  2. If the instrument effect_allele matches the genotype allele_noncoded: flip the coding (beta_ZX *= -1, eaf = 1 - eaf).

  3. If neither direct match is found, try complement alleles (A<->T, G<->C):

    • If the SNP is palindromic (A/T or G/C) and EAF is near 0.5, remove the SNP.

    • If palindromic but EAF clearly differs from 0.5, use EAF to infer strand orientation.

  4. If no match is found even with complements: remove the SNP with reason "allele_mismatch".

References

Hartwig, F. P., Davies, N. M., Hemani, G., & Davey Smith, G. (2016). Two-sample Mendelian randomization: avoiding the downsides of a powerful, widely applicable but potentially fallible technique. International Journal of Epidemiology, 45(6), 1717-1726.

See also

isStrandAmbiguous, getMRInstruments, buildMRCohort

Examples

instruments <- data.frame(
  snp_id = c("rs1", "rs2", "rs3"),
  effect_allele = c("A", "G", "A"),
  other_allele = c("G", "T", "C"),
  beta_ZX = c(0.5, -0.3, 0.2),
  se_ZX = c(0.05, 0.08, 0.06),
  pval_ZX = c(1e-10, 1e-5, 1e-8),
  eaf = c(0.3, 0.45, 0.6),
  stringsAsFactors = FALSE
)
genotypeAlleles <- data.frame(
  snp_id = c("rs1", "rs2", "rs3"),
  allele_coded = c("G", "G", "A"),
  allele_noncoded = c("A", "T", "C"),
  stringsAsFactors = FALSE
)
result <- harmonizeAlleles(instruments, genotypeAlleles)