/bio-causal-genomics-mendelian-randomization

name: bio-causal-genomics-mendelian-randomization
description: Estimate causal effects between exposures and outcomes using genetic variants as instrumental variables with TwoSampleMR. Implements IVW, MR-Egger, weighted median, and MR-PRESSO methods for robust causal inference from GWAS summary statistics. Use when testing whether an exposure causally affects an outcome using genetic instruments.
tool_type: r
primary_tool: TwoSampleMR

Version Compatibility

Reference examples tested with: TwoSampleMR 0.5+, MendelianRandomization 0.9+

Before using code patterns, verify installed versions match. If versions differ:

• R: `packageVersion("<pkg>")` then `?function_name` to verify parameters

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

Mendelian Randomization

**"Test whether my exposure causally affects this outcome using GWAS data"** → Use genetic variants as instrumental variables to estimate causal effects from GWAS summary statistics, applying IVW, MR-Egger, and weighted median methods for robust inference.

• R: `TwoSampleMR::mr()` for multi-method causal estimation
• R: `MendelianRandomization::mr_ivw()` for individual methods

Core Concepts

Mendelian randomization (MR) uses genetic variants as instrumental variables (IVs) to estimate causal effects of exposures on outcomes. Valid instruments must satisfy three assumptions:

1. **Relevance** - The variant is associated with the exposure (F-statistic > 10) 2. **Independence** - The variant is not associated with confounders 3. **Exclusion restriction** - The variant affects the outcome only through the exposure

TwoSampleMR Workflow

**Goal:** Estimate the causal effect of an exposure on an outcome using GWAS summary statistics and genetic instruments.

**Approach:** Extract instruments for the exposure, extract matching outcome data, harmonize allele directions, and run multiple MR methods (IVW, Egger, weighted median, weighted mode).

**"Test if an exposure causally affects an outcome"** -> Use genetic variants as instrumental variables to estimate causal effects from GWAS data.

• R: `TwoSampleMR` (extract_instruments + harmonise_data + mr)
• R: `MendelianRandomization` (mr_input + mr_ivw/mr_egger)

library(TwoSampleMR)

# --- Step 1: Extract instruments for the exposure ---
# From OpenGWAS (requires authentication -- see below)
exposure_dat <- extract_instruments(outcomes = 'ieu-a-2', p1 = 5e-08, clump = TRUE)

# From local GWAS summary statistics
exposure_dat <- read_exposure_data(
  filename = 'exposure_gwas.txt',
  sep = '\t',
  snp_col = 'SNP', beta_col = 'BETA', se_col = 'SE',
  effect_allele_col = 'A1', other_allele_col = 'A2',
  pval_col = 'P', eaf_col = 'EAF'
)

# Clump instruments to remove LD (r2 < 0.001, 10 Mb window)
# r2 < 0.001: Standard threshold to ensure instrument independence
# 10000 kb window: Wide enough to capture long-range LD
exposure_dat <- clump_data(exposure_dat, clump_r2 = 0.001, clump_kb = 10000)

# --- Step 2: Extract outcome data ---
outcome_dat <- extract_outcome_data(snps = exposure_dat$SNP, outcomes = 'ieu-a-7')

# From local summary statistics
outcome_dat <- read_outcome_data(
  filename = 'outcome_gwas.txt',
  sep = '\t',
  snp_col = 'SNP', beta_col = 'BETA', se_col = 'SE',
  effect_allele_col = 'A1', other_allele_col = 'A2',
  pval_col = 'P', eaf_col = 'EAF'
)

# --- Step 3: Harmonize ---
# Ensures effect alleles are aligned between exposure and outcome
dat <- harmonise_data(exposure_dat, outcome_dat, action = 2)

# action = 1: Assume all alleles on forward strand
# action = 2: Try to infer forward strand (default, recommended)
# action = 3: Correct strand for palindromic SNPs using allele frequencies

# --- Step 4: Perform MR ---
results <- mr(dat)

# Run all standard methods
results <- mr(dat, method_list = c(
  'mr_ivw',              # Inverse variance weighted (primary)
  'mr_egger_regression', # MR-Egger (detects pleiotropy)
  'mr_weighted_median',  # Weighted median (robust to 50% invalid)
  'mr_weighted_mode'     # Weighted mode (robust to outliers)
))

OpenGWAS Authentication

OpenGWAS (ieugwasr) requires authentication. The auth system has changed multiple times. Refer to the ieugwasr README for current instructions: https://github.com/MRCIEU/ieugwasr

For reproducibility, prefer downloading GWAS summary statistics directly and using `read_exposure_data()` / `read_outcome_data()` with local files.

Interpreting Results

**Goal:** Evaluate MR evidence through method comparison, heterogeneity testing, and sensitivity analyses.

**Approach:** Compare estimates across methods for consistency, test for heterogeneity (Cochran's Q), pleiotropy (Egger intercept), and single-SNP influence (leave-one-out).

# Method comparison table
results

# Key columns: method, nsnp, b (causal estimate), se, pval
# IVW is the primary method; others are sensitivity analyses
# Consistent direction/magnitude across methods strengthens evidence

# --- Heterogeneity (Cochran's Q) ---
het <- mr_heterogeneity(dat)
# Significant Q-statistic suggests pleiotropy or invalid instruments
# Q p-value < 0.05: Evidence of heterogeneity

# --- Pleiotropy (Egger intercept) ---
pleiotropy <- mr_pleiotropy_test(dat)
# Significant intercept (p < 0.05): Evidence of directional pleiotropy
# Non-significant intercept: No evidence (but low power with few SNPs)

# --- Leave-one-out ---
loo <- mr_leaveoneout(dat)
# Check if any single SNP drives the result
# Causal estimate should remain stable when each SNP is removed

# --- Single SNP analysis ---
single <- mr_singlesnp(dat)

Instrument Strength

**Goal:** Assess whether genetic instruments are strong enough for valid MR inference.

**Approach:** Compute per-instrument F-statistics from exposure effect sizes and standard errors, removing weak instruments (F < 10).

# F-statistic for each instrument
# F = (beta_exposure /