Skip to contents

Multidimensional Mode-Based Estimation for Mendelian Randomization

Source: R/IBMODE.R
IBMODE.Rd

Performs mode-based estimation (MBE) for multiple correlated outcomes simultaneously using multidimensional kernel density estimation. This approach identifies the modal causal effect across multiple traits while accounting for pleiotropy.

Usage

IBMODE(
  BetaXG,
  BetaYG_matrix,
  seBetaXG,
  seBetaYG_matrix,
  phi = c(1, 0.5, 0.25),
  n_boot = 10000,
  alpha = 0.05,
  cov_ratio = NULL,
  ldsc_intercept = NULL,
  seed = NULL
)

Arguments

BetaXG

Numeric vector of SNP-exposure associations

BetaYG_matrix

Matrix of SNP-outcome associations (rows = SNPs, columns = outcomes/traits)

seBetaXG

Numeric vector of standard errors for BetaXG

seBetaYG_matrix

Matrix of standard errors for BetaYG_matrix

phi

Numeric vector of bandwidth multipliers for sensitivity analysis (default: c(1, 0.5, 0.25))

n_boot

Integer, number of bootstrap iterations for standard error estimation (default: 10000)

alpha

Numeric, significance level for confidence intervals (default: 0.05)

cov_ratio

Optional numeric vector of per-instrument cross-trait covariances of the ratio estimates (\(\sigma_{12,k}\)); when supplied (or derived from ldsc_intercept), the bootstrap draws each instrument's outcome pair jointly from a bivariate normal, accounting for correlation induced by overlapping outcome GWAS samples. NULL (default) or all-zero gives the independent bootstrap. Two-outcome analyses only.

ldsc_intercept

Optional scalar cross-trait LD-score regression intercept for the (primary, auxiliary) pair; if cov_ratio is NULL and this is supplied, cov_ratio is formed internally.

seed

Optional integer random seed for a reproducible bootstrap (default: NULL).

Value

A data frame containing:

Method

Method name

phi

Bandwidth multiplier used

Estimate_

Causal effect estimates for each outcome

SE_

Standard errors for each outcome

CI_low_

Lower confidence interval bounds

CI_upp_

Upper confidence interval bounds

P_

P-values for each outcome

Details

The function uses weighted multidimensional kernel density estimation to find the mode of the joint distribution of causal effects across multiple outcomes. Bootstrap resampling is used to estimate standard errors; when the primary and auxiliary outcome GWAS share samples, supply cov_ratio (or ldsc_intercept) so the bootstrap is drawn from the appropriate bivariate distribution. Multiple bandwidth parameters (phi) can be tested for sensitivity analysis.