Metabolomics Group Comparison Statistics

metabolomics-group-comparison-statistics

Summary

Compute differential abundance statistics (p-values, adjusted p-values, log2 fold changes) from raw metabolomics data matrices by comparing two experimental groups. This skill produces summary statistics that serve as input for downstream pathway enrichment analysis.

When to use

You have a metabolomics matrix with metabolites as rows and samples as columns, each sample labeled with one of two experimental group identifiers (e.g., 'TK-CMV' or 'K-CMV'), and you need to identify which metabolites differ significantly between the groups before performing enrichment analysis.

When NOT to use

• Input is already a precomputed summary statistics table with p-values and fold changes — skip to enrichment analysis directly.
• Sample sizes are extremely small (< 3 replicates per group) — statistical power will be insufficient to detect true differences.
• Metabolites have been heavily filtered or preprocessed independently by group — violates assumption of computing statistics on the full raw matrix.

Inputs

• Metabolomics data matrix (metabolites × samples) with numeric abundance values
• Sample metadata table with group assignments ('TK-CMV', 'K-CMV', or equivalent binary group labels)
• Metabolite identifier annotations (e.g., KEGG IDs, m/z values, retention times)

Outputs

• Differential analysis results table (data.frame) with columns: metabolite_id, p_value, adjusted_p_value, log2_fold_change
• Summary statistics table suitable for downstream enrichment analysis

How to apply

Load the metabolomics data matrix ensuring metabolites are rows and samples are columns, with sample metadata including group assignment. Call the run_de() function, which performs statistical testing (typically t-test or Mann-Whitney U) to generate p-values for each metabolite. The function also computes adjusted p-values (using methods like Benjamini-Hochberg) and log2 fold changes comparing mean abundance between groups. Filter metabolites using a p-value cutoff (commonly p < 0.05) to retain statistically significant results. Output the differential analysis results table containing metabolite identifiers, p-values, adjusted p-values, and log2 fold changes for input to enrichment analysis.

Related tools

• run_de() (Performs differential abundance analysis comparing two groups, generating p-values, adjusted p-values, and log2 fold changes for each metabolite) — https://github.com/biodatalab/enrichmet
• R (Statistical computing environment in which run_de() is executed and metabolomics matrices are manipulated)

Examples

da_out <- run_de(inputMetabolites = metabolite_matrix, groups = sample_groups); results <- enrichmet(inputMetabolites = NULL, PathwayVsMetabolites = PathwayVsMetabolites, da_results = da_out, p_value_cutoff = 0.05)

Evaluation signals

• Output table has one row per metabolite and contains no missing values in p_value, adjusted_p_value, and log2_fold_change columns.
• P-values and adjusted p-values fall within [0, 1]; adjusted p-values are ≥ corresponding raw p-values (monotonicity check).
• Log2 fold changes reflect the direction and magnitude of group difference; metabolites with significant p-values (p < 0.05) show non-zero fold changes.
• Row count of output matches input metabolite count (or is less only if filtering was applied with explicit criteria).
• Output is compatible with enrichmet() function input format (da_results parameter accepts the table schema directly).

Limitations

• run_de() assumes data are suitable for parametric or non-parametric tests; highly zero-inflated or bimodal distributions may require specialized methods (e.g., zero-inflated models, compositional analysis).
• Statistical power depends critically on sample size and biological variance; small sample sizes (n < 5 per group) yield high false discovery rates.
• Adjusted p-values are sensitive to the multiple-testing correction method; Benjamini-Hochberg is standard but may be conservative if many true signals are present.
• The function requires group labels to be binary or explicitly specified; multi-group comparisons require pairwise post-hoc testing.
• No changelog is available for the enrichmet package, making it difficult to track changes to run_de() behavior across versions.

Evidence

• [other] Load the raw metabolomics data matrix with metabolites as rows and samples as columns, ensuring samples are labeled with group assignments ('TK-CMV' or 'K-CMV').: "Load the raw metabolomics data matrix with metabolites as rows and samples as columns, ensuring samples are labeled with group assignments ('TK-CMV' or 'K-CMV')"
• [other] Call run_de() function to perform differential analysis comparing the two groups, generating summary statistics including p-values, adjusted p-values, and log2 fold changes for each metabolite.: "Call run_de() function to perform differential analysis comparing the two groups, generating summary statistics including p-values, adjusted p-values, and log2 fold changes for each metabolite"
• [other] The run_de() function accepts a metabolomics matrix with metabolites as rows and samples as columns, along with group labels ('TK-CMV' and 'K-CMV'), and produces differential analysis output containing p-values, adjusted p-values, and log2 fold changes that serve as input for downstream enrichment analysis.: "The run_de() function accepts a metabolomics matrix with metabolites as rows and samples as columns, along with group labels ('TK-CMV' and 'K-CMV'), and produces differential analysis output"
• [intro] the workflow should begin with the run_de() function. This step generates the summary statistics required for enrichment, including p-values, adjusted p-values, and log2 fold changes.: "the workflow should begin with the run_de() function. This step generates the summary statistics required for enrichment, including p-values, adjusted p-values, and log2 fold changes"
• [intro] Filter metabolites using p_value_cutoff parameter [section=intro; evidence='p_value_cutoff = 0.05']: "p_value_cutoff = 0.05"