OmicVerse Bulk Metabolomics — Untargeted LC-MS & Lipidomics

Goal

Take an LC-MS peak table (m/z + RT-coded features) or a lipidomics matrix (LIPID MAPS shorthand var-names) into an AnnData, then run the two ID-aware downstream paths — peak annotation + mummichog pathway inference for untargeted metabolomics, lipid-class parsing + class aggregation + LION enrichment for lipidomics. The output is a list of pathway / ontology hits backed by the empirical feature p-values, plus a class-aggregated lipid matrix when requested.

This skill assumes the upstream preprocessing chain (impute / normalize / log) has already run; for the chain see the omicverse-bulk-metabol-preprocessing skill. ID mapping for named metabolites (HMDB / KEGG / ChEBI ID resolution) and MSEA pathway enrichment are in the pathway-multifactor skill.

Quick Workflow

Untargeted LC-MS path (t_metabol_04)

1. Load the LC-MS peak table with ov.metabol.read_lcms(..., feature_id_sep=...) so m/z and RT get parsed into var['m_z'] and var['rt'].
2. Apply preprocessing (PQN + log) — don't impute zeros for LC-MS (zeros = below-detection).
3. Run differential to get a per-peak p-value vector.
4. Untargeted LC-MS often has thousands of peaks vs. low n — BH-FDR is too strict to discriminate. For the volcano, pass use_pvalue=True and clip the log2fc axis to keep below-detection extremes from blowing the figure.
5. Enrich pathways from m/z + p-value with ov.metabol.mummichog_basic(...) — a permutation-based pathway test that handles the FDR burden internally.
6. Optionally inspect candidate compound annotations per peak with ov.metabol.annotate_peaks(...) — adduct-resolved KEGG candidates by ppm tolerance.
7. For sanity checks: build a synthetic m/z set from a known KEGG pathway (e.g. via fetch_chebi_compounds() + load_pathways()), seed it with low p-values, and confirm mummichog recovers the seeded pathway.