AGENT SKILL

bio-causal-genomics-mediation-analysis

Decompose genetic effects into direct and indirect paths through mediating variables using the mediation R package. Tests whether gene expression, methylation, or other molecular phenotypes mediate the effect of genetic variants on disease. Use when testing whether a molecular...

License 许可
MIT
最近更新
2026-10-12
Stars
★ 2
Forks
0

01 前言 Front Matter

name
bio-causal-genomics-mediation-analysis
author
AIPOCH
license
MIT
description
Decompose genetic effects into direct and indirect paths through mediating variables using the mediation R package. Tests whether gene expression, methylation, or other molecular phenotypes mediate the effect of genetic variants on disease. Use when testing whether a molecular...

02 资源文件 2

03 SKILL.md 正文

Source: https://github.com/aipoch/medical-research-skills

Version Compatibility

Reference examples tested with: R stats (base), ggplot2 3.5+

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.

Mediation Analysis

"Test whether gene expression mediates the effect of this variant on disease" → Decompose the total genetic effect into direct and indirect (mediated) paths through a molecular phenotype, estimating ACME, ADE, and proportion mediated with bootstrap confidence intervals.

  • R: mediation::mediate() for causal mediation analysis

Framework

Causal mediation decomposes the total effect of a treatment (genotype) on an outcome (phenotype) into:

  • ACME (Average Causal Mediation Effect) - Indirect effect through the mediator
  • ADE (Average Direct Effect) - Direct effect not through the mediator
  • Total effect = ACME + ADE
  • Proportion mediated = ACME / Total effect

Typical genomic applications:

  • SNP -> gene expression (mediator) -> disease
  • SNP -> DNA methylation (mediator) -> gene expression
  • SNP -> protein levels (mediator) -> clinical outcome

Basic Mediation with the mediation Package

Goal: Decompose a genetic effect into direct and indirect (mediated) paths through a molecular phenotype.

Approach: Fit separate models for mediator and outcome, then run mediate() with bootstrap to estimate ACME (indirect), ADE (direct), and proportion mediated.

library(mediation)

# --- Step 1: Fit mediator model ---
# How does the treatment (genotype) affect the mediator (expression)?
mediator_model <- lm(expression ~ genotype + age + sex + pc1 + pc2, data = dat)

# --- Step 2: Fit outcome model ---
# How do treatment and mediator jointly affect the outcome?
# For binary outcome, use glm with family = binomial
outcome_model <- glm(
  disease ~ genotype + expression + age + sex + pc1 + pc2,
  data = dat, family = binomial
)

# --- Step 3: Run mediation analysis ---
# treat: name of treatment variable (genotype)
# mediator: name of mediator variable (expression)
# boot = TRUE: Use nonparametric bootstrap for CIs
# sims: Number of bootstrap simulations (1000 minimum for publication)
med_result <- mediate(
  mediator_model, outcome_model,
  treat = 'genotype', mediator = 'expression',
  boot = TRUE, sims = 1000
)

summary(med_result)
# Key outputs:
# ACME: Indirect effect (through expression)
# ADE: Direct effect (not through expression)
# Total Effect: ACME + ADE
# Prop. Mediated: ACME / Total

Interpreting Results

# Extract key quantities
acme <- med_result$d0           # Indirect (mediated) effect
acme_ci <- med_result$d0.ci     # 95% CI for ACME
ade <- med_result$z0            # Direct effect
total <- med_result$tau.coef    # Total effect
prop_med <- med_result$n0       # Proportion mediated

cat('ACME (indirect):', round(acme, 4), '\n')
cat('ACME 95% CI:', round(acme_ci[1], 4), 'to', round(acme_ci[2], 4), '\n')
cat('ADE (direct):', round(ade, 4), '\n')
cat('Total effect:', round(total, 4), '\n')
cat('Proportion mediated:', round(prop_med, 3), '\n')

# Significant ACME (CI excludes 0): Evidence for mediation
# Proportion mediated > 0.2: Meaningful mediation
# Proportion mediated > 0.8: Mediator explains most of the effect

eQTL Mediation

Goal: Test whether gene expression mediates the effect of an eQTL on a disease outcome across multiple genes.

Approach: Wrap the mediation workflow in a function, loop over candidate genes, and adjust p-values for multiple testing.

library(mediation)

run_eqtl_mediation <- function(dat, snp_col, expr_col, outcome_col, covariates) {
  covar_formula <- paste(covariates, collapse = ' + ')

  med_formula <- as.formula(paste(expr_col, '~', snp_col, '+', covar_formula))
  out_formula <- as.formula(paste(outcome_col, '~', snp_col, '+', expr_col, '+', covar_formula))

  med_model <- lm(med_formula, data = dat)

  if (length(unique(dat[[outcome_col]])) == 2) {
    out_model <- glm(out_formula, data = dat, family = binomial)
  } else {
    out_model <- lm(out_formula, data = dat)
  }

  result <- mediate(
    med_model, out_model,
    treat = snp_col, mediator = expr_col,
    boot = TRUE, sims = 1000
  )

  data.frame(
    snp = snp_col, gene = expr_col,
    acme = result$d0, acme_p = result$d0.p,
    ade = result$z0, ade_p = result$z0.p,
    total = result$tau.coef, total_p = result$tau.p,
    prop_mediated = result$n0
  )
}

# Example: test mediation for multiple genes
genes <- c('GENE_A', 'GENE_B', 'GENE_C')
covars <- c('age', 'sex', 'pc1', 'pc2', 'pc3')

mediation_results <- do.call(rbind, lapply(genes, function(g) {
  run_eqtl_mediation(dat, 'rs12345', g, 'disease_status', covars)
}))

# Adjust for multiple testing
mediation_results$acme_fdr <- p.adjust(mediation_results$acme_p, method = 'BH')

Multi-Omics Mediation

Goal: Test cascading mediation chains across multiple molecular layers (e.g., SNP -> methylation -> expression -> disease).

Approach: Fit sequential models for each link in the chain and run separate mediation analyses for each mediator-outcome pair.

# Test mediation chains: SNP -> methylation -> expression -> disease
library(mediation)

# Step 1: SNP -> methylation
mod_meth <- lm(methylation ~ genotype + age + sex, data = dat)

# Step 2: methylation -> expression (controlling for genotype)
mod_expr <- lm(expression ~ methylation + genotype + age + sex, data = dat)

# Step 3: expression -> disease (controlling for methylation and genotype)
mod_disease <- glm(
  disease ~ expression + methylation + genotype + age + sex,
  data = dat, family = binomial
)

# Test methylation as mediator of SNP -> expression
med_meth_expr <- mediate(mod_meth, mod_expr, treat = 'genotype', mediator = 'methylation',
                         boot = TRUE, sims = 1000)

# Test expression as mediator of methylation -> disease
med_expr_disease <- mediate(mod_expr, mod_disease, treat = 'methylation', mediator = 'expression',
                            boot = TRUE, sims = 1000)

High-Dimensional Mediation (HDMA)

Goal: Test thousands of potential mediators simultaneously (e.g., all CpG sites) to identify which mediate a genetic effect.

Approach: Use HIMA's penalized regression to jointly select significant mediators from a high-dimensional mediator matrix and estimate their indirect effects.

# For testing many potential mediators simultaneously (e.g., all CpG sites)
# install.packages('HIMA')
library(HIMA)

# X: treatment (genotype), M: high-dimensional mediators, Y: outcome
# HIMA uses penalized regression to select significant mediators

result <- hima(
  X = dat$genotype,
  Y = dat$disease,
  M = as.matrix(dat[, mediator_cols]),
  COV.XM = as.matrix(dat[, covariate_cols]),
  Y.family = 'binomial',
  M.family = 'gaussian',
  penalty = 'MCP'    # Minimax concave penalty (default)
)

# Results: significant mediators with estimated indirect effects
significant_mediators <- result[result$BH.FDR < 0.05, ]

Assumptions and Diagnostics

# --- Sequential ignorability assumption ---
# 1. No unmeasured confounders between treatment and mediator
# 2. No unmeasured confounders between mediator and outcome
# 3. No unmeasured confounders between treatment and outcome
# This assumption is UNTESTABLE but can be probed with sensitivity analysis

# --- Sensitivity analysis ---
# Tests how robust results are to unmeasured confounding
sens <- medsens(med_result, rho.by = 0.1, effect.type = 'indirect', sims = 1000)
summary(sens)

# rho: Correlation between residuals of mediator and outcome models
# At what rho does ACME cross zero? (larger |rho| = more robust)
# rho at which ACME = 0 is called the sensitivity parameter
# |rho| > 0.3: Reasonably robust to unmeasured confounding

plot(sens)

Visualization

library(ggplot2)

plot_mediation_diagram <- function(acme, ade, total, prop_med) {
  cat('Mediation Path Diagram:\n\n')
  cat('  Genotype ---[a]---> Mediator ---[b]---> Outcome\n')
  cat('      |                                     ^\n')
  cat('      +----------[c\' (ADE)]----------------+\n')
  cat('\n')
  cat('  Indirect (a*b = ACME):', round(acme, 4), '\n')
  cat('  Direct (c\' = ADE):', round(ade, 4), '\n')
  cat('  Total (c):', round(total, 4), '\n')
  cat('  Proportion mediated:', round(prop_med, 3), '\n')
}

plot_mediation_results <- function(results_df) {
  results_df$gene <- factor(results_df$gene, levels = results_df$gene[order(results_df$prop_mediated)])

  ggplot(results_df, aes(x = gene, y = prop_mediated)) +
    geom_col(fill = 'steelblue', alpha = 0.7) +
    geom_hline(yintercept = 0.2, linetype = 'dashed', color = 'red', alpha = 0.5) +
    coord_flip() +
    labs(x = NULL, y = 'Proportion Mediated', title = 'Mediation by Gene Expression') +
    theme_minimal()
}

Related Skills

  • mendelian-randomization - Causal inference using genetic instruments
  • colocalization-analysis - Test if signals share a causal variant
  • population-genetics/association-testing - GWAS for treatment-outcome associations
  • multi-omics-integration/mofa-integration - Multi-omics data for mediation chains

Input Validation

This skill accepts requests that match the documented purpose of bio-causal-genomics-mediation-analysis and include enough context to complete the workflow safely.

Do not continue the workflow when the request is out of scope, missing a critical input, or would require unsupported assumptions. Instead respond:

bio-causal-genomics-mediation-analysis only handles its documented workflow. Please provide the missing required inputs or switch to a more suitable skill.

04 同仓库更多 Skills

mrsonord2240
mrsonord2240/openscience-specialists

bio-workflows-metabolomics-pipeline

Orchestrates the untargeted LC-MS metabolomics pipeline end-to-end (xcms 4.x feature extraction, QC/drift/normalization, confidence-stratified annotation, permutation-validated statistics, background-aware pathway mapping), naming what each stage decides and where it silently fails. Use when running a full LC-MS metabolomics study from raw mzML to enriched pathways and needing the honest handoffs between stages. Each stage defers to its component skill for parameters and traps; for stable-isotope flux (a separate branch, not this untargeted flow) see metabolomics/isotope-tracing.

★ 2
mrsonord2240
mrsonord2240/openscience-specialists

bio-metabolomics-xcms-preprocessing

Programmatic untargeted LC-MS feature extraction in R with the modern xcms 4.x MsExperiment/XcmsExperiment API, taking raw mzML to a feature table via CentWave peak detection, retention-time alignment, peak-density correspondence, gap-filling, CAMERA redundancy collapse, and built-in QC feature filtering. Use when converting centroided LC-MS runs into a features-by-samples matrix and deciding centWave/grouping/alignment parameters. For drift correction and QC/CV filtering execution see metabolomics/normalization-qc; for metabolite identification see metabolomics/metabolite-annotation; for the MS-DIAL GUI alternative with MS2Dec deconvolution see metabolomics/msdial-preprocessing; for downstream statistics see metabolomics/statistical-analysis.

★ 2
mrsonord2240
mrsonord2240/openscience-specialists

bio-metabolomics-statistical-analysis

Decision-grade statistical analysis for metabolomics intensity tables. Covers transformation and scaling (Pareto vs unit-variance as a hidden hypothesis), unsupervised structure (PCA/HCA for QC), permutation-validated PLS-DA/OPLS-DA (R2 vs Q2, double CV, VIP as heuristic), univariate testing (Welch/Mann-Whitney/ANOVA/LMM with covariate adjustment), and dependence-aware multiple testing. Use when testing which metabolites differ, building or validating a discriminant model, choosing a scaling, or correcting many correlated tests. For sample-wise normalization/drift correction see metabolomics/normalization-qc; for ML classifiers and selection-inside-CV leakage see machine-learning/biomarker-discovery and machine-learning/model-validation; for pathway interpretation see metabolomics/pathway-mapping; for design/power/multiplicity regime see experimental-design/multiple-testing.

★ 2
mrsonord2240
mrsonord2240/openscience-specialists

bio-metabolomics-pathway-mapping

Maps metabolomics results to biological pathways via over-representation (ORA), metabolite-set enrichment (MSEA/QEA), mummichog/PSEA on raw m/z peaks, and network-diffusion enrichment (FELLA), with correct background-set construction and honest interpretive ceilings. Use when interpreting differential metabolites or an untargeted LC-MS feature table in pathway context, choosing ORA vs MSEA vs mummichog vs topology, or setting the reference/background set. For annotation confidence levels feeding ORA see metabolomics/metabolite-annotation; for gene-set concepts see pathway-analysis/go-enrichment and pathway-analysis/gsea; for joint gene+metabolite pathways see multi-omics-integration/mofa-integration.

★ 2
mrsonord2240
mrsonord2240/openscience-specialists

bio-metabolomics-normalization-qc

Designs QC, corrects signal drift, removes batch effects, filters features, normalizes samples, and imputes missing values for untargeted LC-MS/GC-MS metabolomics, framing each step as a measurement model that can create or erase biological signal. Use when processing a peak/feature table before statistical analysis, choosing a drift-correction or sample-normalization method, deciding QC RSD vs D-ratio filtering, or handling left-censored missing values. The feature table is produced by metabolomics/xcms-preprocessing or metabolomics/msdial-preprocessing; transformation/scaling for modeling defers to metabolomics/statistical-analysis; cross-study design issues link to experimental-design/batch-design.

★ 2
mrsonord2240
mrsonord2240/openscience-specialists

bio-metabolomics-msdial-preprocessing

Runs the MS-DIAL preprocessing workflow (peak picking, MS2Dec spectral deconvolution, alignment, gap-filling) and imports the alignment-result table into R or Python with honest filtering. Use when preprocessing LC-MS DDA/DIA (SWATH) or GC-MS raw data with MS-DIAL, deciding MS-DIAL vs XCMS, configuring the MSDIALCUI console run, or parsing an MS-DIAL export into a clean feature matrix. For programmatic R peak detection and the feature-table-as-artifact framing see metabolomics/xcms-preprocessing; for lipid annotation mode see metabolomics/lipidomics; for MSI-level confidence honesty see metabolomics/metabolite-annotation; for drift correction and QC see metabolomics/normalization-qc.

★ 2