Untargeted Metabolomics Feature Interpretation

Reconstruct metabolic-network-based functional activity prediction from a feature table

Summary

This skill predicts functional pathway activity directly from untargeted metabolomics feature tables (m/z, retention time, intensity) by mapping features to a metabolic network and propagating activity signals through biochemical pathways, without requiring prior metabolite identification. It enables rapid functional interpretation of mass spectrometry data via network-based inference.

When to use

Apply this skill when you have an untargeted metabolomics feature table (m/z values, retention times, p-values from statistical testing) and need to infer which metabolic pathways are active without performing metabolite identification. Use it especially when you have feature-level p-values from differential analysis but lack confident metabolite annotations, or when you want pathway-level functional interpretation that is robust to annotation uncertainty.

When NOT to use

• Your metabolic model lacks chemical formulas for compounds and no identifier translation module is available to look them up
• You have already performed comprehensive metabolite identification and want to use those identities directly in quantitative pathway analysis rather than network-based inference
• Your feature table contains no statistical test results (p-values) or intensity values to propagate as activity signals through the network

Inputs

• Untargeted metabolomics feature table (TSV/CSV with m/z, retention time, intensity, p-value columns)
• Optional: Feature annotation table (from authentic standards and MS/MS, structured in metDataModel format)
• Metabolic model in JSON format (list of pathways, reactions, compounds with chemical formulas; default models provided by mummichog or user-supplied)
• Optional: Compound identifier translation mapping

Outputs

• Result tables and figures (HTML report as main deliverable)
• JSON strings of pathway analysis results (for programmatic use)
• JSON strings of network module analysis results
• Ranked list of predicted functional pathways with activity scores

How to apply

Load the untargeted metabolomics feature table containing m/z, retention time, and intensity measurements into Python. Construct or load a metabolic network representation (provided by mummichog or user-supplied in JSON format) that captures biochemical pathway organization and metabolite connectivity, ensuring compound formulas are present or can be looked up via identifiers. Map feature m/z values to network nodes using mass-matching (without explicit metabolite identification), accounting for adducts and neutral masses. Propagate activity signals (derived from feature p-values or intensities) through the metabolic network using graph-based algorithms to infer which pathways show coordinated activation. Aggregate and rank pathway-level predictions by network evidence strength, outputting structured results (pathway activity scores, ranked functional predictions) in HTML and JSON formats for both visual inspection and programmatic downstream use.

Related tools

• mummichog (Core Python package that implements network-based pathway prediction from untargeted metabolomics feature tables; performs mass-matching, network propagation, and pathway ranking) — https://github.com/metabolomics-cloud/mummichog
• metDataModel (Data model framework for structuring metabolomics annotations (used by mummichog to organize feature metadata))
• JMS (Performs MS/MS-based metabolite annotation (optional input to enrich feature annotations before network analysis)) — https://github.com/shuzhao-li-lab/JMS
• mass2chem (Utility library for mass spectrometry data interpretation and adduct calculations (underpins m/z-to-compound mapping))

Examples

python3 -m mummichog.main -i tests/ineurons_ttest_1127.tsv -j testneuron -a tests/empCpds_with_annotations.json -d .

Evaluation signals

• Output JSON contains non-empty lists of predicted pathways ranked by evidence strength (network connectivity and signal propagation depth)
• Predicted pathway scores correlate with expected biology (known differential metabolic activity in the experimental context should rank highly)
• Feature-to-network mappings respect chemical mass tolerance (adducts and neutral masses match observed m/z within instrument accuracy)
• HTML report contains both pathway rankings and network visualizations, confirming signal propagation through biochemically connected nodes
• Results remain stable across multiple runs with same inputs, indicating deterministic network propagation

Limitations

• Accuracy depends critically on the completeness and correctness of the input metabolic model; incomplete or incorrect pathway definitions will produce misleading rankings
• Requires chemical formulas for compounds (either in model or via lookup); models missing this information cannot perform accurate mass-matching
• Pathway definition may be unavailable in some metabolic models; 'subsystem' annotation can be used as a substitute but may not represent true biological pathways
• Network-based inference cannot distinguish between direct pathway activation and indirect effects propagated through network connectivity
• Results do not identify individual metabolites; functional predictions operate at pathway level and should be validated with targeted metabolite analysis

Evidence

• [intro] Feature-table-based pathway prediction without metabolite ID: "It leverages the organization of metabolic networks to predict functional activity directly from feature tables, bypassing metabolite identification."
• [other] Workflow steps for network mapping and signal propagation: "Map feature m/z values to network nodes via mass-matching without performing explicit metabolite identification. Propagate activity signals through the metabolic network using graph-based algorithms"
• [readme] Input format specification: "User supplied features with m/z, rtime, p-value from a statistical test. If no unique feature ID is supplied, row_number will be used as ID."
• [readme] Metabolic model requirements and format: "The format of a metabolic model contains list_of_pathways, list_of_reactions and list_of_compounds. We need chemical formulas of the compounds, which may not be available in a GSMM."
• [readme] Output format and deliverables: "Outpout are 1. Result tables and figures, result.html as ver 2. 2. JSON strings from pathway analysis and network module analysis for programmatic use."
• [readme] Chemical formula and adduct handling: "If compound formulas are provided in a model, be sure to have correct neutral mass and not a salt format. Salts are formed in both biological systems and in mass spectrometry. We use neutral formula"