Multiple Hypothesis Correction In Metabolomics

multiple-hypothesis-correction-in-metabolomics

Summary

Apply adjusted p-value thresholds and multiple-testing corrections during metabolite set enrichment analysis to control false discovery rates across pathway and metabolite hypotheses. This prevents spurious pathway assignments when testing many metabolites and pathways simultaneously.

When to use

After computing raw p-values from differential metabolomics analysis or Fisher's exact test enrichment, and before reporting significant pathways or metabolites. Essential when testing multiple metabolites against multiple pathway hypotheses (e.g., 20+ metabolites × 50+ pathways), where uncorrected p-values inflate Type I error.

When NOT to use

• Input is a single metabolite or single pathway (no multiple hypothesis problem)
• Study design involves only a priori hypotheses without exploratory testing (hypothesis-driven studies with pre-registered comparisons may use less stringent correction)
• Raw p-values are already adjusted for multiple testing by the upstream differential analysis tool

Inputs

• Raw p-values and log2 fold changes from differential analysis
• MetSEA results table with p-values and NES (Normalized Enrichment Scores)
• Pathway-to-metabolite mapping (PathwayVsMetabolites reference file)
• Metabolite count and pathway occurrence matrices

Outputs

• Filtered MetSEA results table (pathways passing padj threshold)
• Filtered pathway enrichment table with significant pathways only
• Filtered metabolite centrality table (metabolites meeting co-occurrence cutoff)
• Summary statistics: number of pathways tested vs. passed, number of metabolites retained

How to apply

During enrichment analysis, apply a p-value cutoff (e.g., padj < 0.05 or p < 0.05) and optional adjusted p-value filtering to control family-wise error rate or false discovery rate. The enrichmet() function filters metabolites using p_value_cutoff and applies adjusted p-value thresholds to MetSEA results before extracting significant pathway associations. Set min_pathway_occurrence and min_metabolite_occurrence parameters to enforce minimum co-occurrence thresholds, further reducing spurious pathway-metabolite pairs. Use fgsea's internal FDR correction (padj column) to identify Normalized Enrichment Scores significant at FDR < 0.05. Document all filtering thresholds in the methods section so results are reproducible and comparable across studies.

Related tools

• fgsea (Computes Normalized Enrichment Scores with internal FDR correction (padj column) for metabolite set enrichment, automatically controlling false discovery across pathways)
• enrichmet (Wrapper function integrating fgsea and Fisher's exact test with user-configurable p_value_cutoff, min_pathway_occurrence, and min_metabolite_occurrence parameters to filter results downstream of enrichment computation) — https://github.com/biodatalab/enrichmet
• R (Statistical environment for implementing multiple-testing correction logic and filtering pipelines)

Examples

results <- enrichmet(inputMetabolites = inputMetabolites, PathwayVsMetabolites = PathwayVsMetabolites, da_results = da_out, p_value_cutoff = 0.05, min_pathway_occurrence = 2, min_metabolite_occurrence = 1)

Evaluation signals

• Verify p-value cutoff was applied consistently: count(pathways where padj < threshold) ≤ count(pathways where padj < 0.1 or p < cutoff); confirm no pathway with padj > threshold appears in final results table
• Confirm min_pathway_occurrence and min_metabolite_occurrence filters eliminated spurious low-frequency associations: check that all retained metabolite-pathway pairs appear in at least the specified minimum number of pathways/metabolites
• Check that MetSEA results show FDR-corrected significance: inspect results$MetSEA table for padj column; confirm only NES values with padj < 0.05 are highlighted in plots
• Reproducibility check: re-run enrichmet() with identical p_value_cutoff, min_pathway_occurrence, and min_metabolite_occurrence parameters and verify identical filtered results; compare row counts before/after filtering against documented thresholds
• Report effect of correction: document in methods how many raw hypotheses were tested (total metabolites × total pathways) vs. how many passed filtering; compute and report the effective correction factor (e.g., Bonferroni-equivalent)

Limitations

• Conservative correction (e.g., Bonferroni) may reduce statistical power in high-dimensional metabolomics; FDR control (via fgsea padj) is generally preferred but depends on having sufficient samples and metabolites to estimate the null distribution accurately
• Thresholds (p_value_cutoff, min_pathway_occurrence) are data- and study-dependent; no universal defaults; setting too stringent thresholds may exclude true discoveries, while too lenient thresholds reintroduce false positives
• Multiple-testing correction assumes independence between metabolites and pathways; if metabolites are highly correlated or pathways heavily overlap, correction may be overly conservative or ineffective
• Adjusted p-values (padj) are only valid within the set of hypotheses tested; if new metabolites or pathways are added post-hoc, p-values must be recomputed

Evidence

• [intro] p_value_cutoff parameter controls significance filtering in enrichment analysis: "Filter metabolites using p_value_cutoff parameter [section=intro; evidence='p_value_cutoff = 0.05']"
• [intro] fgsea provides FDR-corrected p-values (padj) for metabolite set enrichment: "enrichmet integrates fgsea for fast MetSEA, igraph for topology-based metrics, and curated KEGG data for enrichment using Fisher's Exact Test—all accessible via a single function call."
• [intro] min_pathway_occurrence and min_metabolite_occurrence enforce co-occurrence thresholds to reduce spurious associations: "Filter pathways using min_pathway_occurrence parameter [section=intro; evidence='min_pathway_occurrence = 2']; Filter metabolites using min_metabolite_occurrence parameter [section=intro;"
• [other] MetSEA results table includes p-values and adjusted p-values for significance assessment: "Extract the MetSEA results table containing pathway names, NES values, p-values, and adjusted p-values."
• [readme] README demonstrates filtering applied during enrichment with configurable cutoff parameters: "p_value_cutoff = 0.1, # Use a reasonable cutoff"