Build a metabolomics panel from biology, not catalogue.

Search Reactome pathways (e.g. "tca cycle", "bile acid", "sars-cov-2"). Adds every metabolite participating in any reaction of the chosen pathways as a candidate.

If you're running paired microbiome + metabolomics, add taxa you expect to increase and taxa you expect to decrease in the case group. Skip both fields if you're not doing microbiome — the other streams cover you.

Three anchors run together: direct (disease → gene → reaction → metabolite) finds enzymatic links; pathway-expansion harvests every Reactome pathway the disease's genes participate in (signaling, regulatory, immune) and surfaces metabolites from those pathways; literature anchor queries NCBI PubMed for chemical-MeSH terms co-cited with the disease in metabolomics / biomarker studies. The result table tags each metabolite with which anchors reached it (direct · pathway_expand · literature·pubmed). First lookup per disease takes ~2–5 s (NCBI), then cached. Autocomplete shows any MONDO/Orphanet term — even ones without graph edges, since literature can still anchor them.

First request after a worker restart may take up to ~30 s while the GIZMO graph loads into memory; subsequent runs are fast.