GIZMO is a calibrated multi-omics knowledge graph. Nodes span metabolites, reactions, genes, transcripts, proteins, phenotypes, and diseases, drawn from the same authoritative sources (Reactome, MetaCyc, Ensembl, UniProt, MONDO, HPO, and more) and linked through edges whose couplings are tuned on held-out biological triples. Feed it hits from any omics layer and it returns ranked diseases, causal chains, and druggable targets, with posterior confidence a chemist or clinician can actually read.
Map your hit list onto the graph: metabolites by HMDB / KEGG / PubChem, genes and transcripts by Ensembl / Entrez, proteins by UniProt — with fuzzy name matching and dataset-memory from prior corrections. Anchors pin reference nodes so real biomarkers don't drift under graph updates.
Classical hypergeometric ORA against Reactome pathways with per-pathway fold enrichment and BH-FDR. Use your list as the foreground; an informed background universe (measured HMDBs, or community-derived) keeps the test honest.
Tie your signatures to clinical outcomes. For every signature, GIZMO reconstructs the most-probable metabolite → reaction → gene → phenotype path. Each edge carries a learned coupling; each chain ends with a posterior probability that's calibrated against held-out triples.
Click any reaction on a results page, simulate its removal, and re-score every downstream disease and target. Reactions rank by |Δscore| — the nodes whose deletion most changes the graph's disease associations are the highest-leverage interventions.
Genes carrying high posterior probability in the causal chains are scored against tractability and clinical-candidate signals. Output will be a short list ready for medicinal-chemistry triage.
Edge-type couplings are tuned by coordinate descent against held-out data triples. GIZMO ships a benchmark suite comparing default vs tuned couplings, and comparing GIZMO (heuristic + BP) against ORA-only and random baselines. Every claim is reproducible against the pinned graph snapshot.
One example of the general capability. Input: a list of differentially abundant plasma metabolites from a published AD vs control cohort. The same framing applies to a list of differentially expressed genes or proteins from a matched transcriptomics / proteomics study GIZMO resolves either kind of signature to graph nodes, propagates evidence, and returns a ranked disease list alongside the causal chains supporting the top hits.
Those tools rank pathways by over-representation. GIZMO ranks diseases by calibrated graph-propagated probability, then traces each ranking back to a specific causal chain you can audit. A pathway you'd already suspected is reassurance; the unexpected genes at high posterior are the lead the analysis exists to produce.
Graph is rebuilt additively — new data layers extend existing nodes rather than overwriting them — so every production run is anchored to a single named active graph with a version stamp.
| Capability | GIZMO | MetaboAnalyst | IMPaLA | OmicsNet | Open Targets |
|---|---|---|---|---|---|
| Metabolite ORA (hypergeometric, FDR-corrected) | ✓ | ✓ | ✓ (joint with transcriptomics) | partial | — |
| Disease ranking from a metabolite list | ✓ Bayesian propagation | pathway → disease indirect | — | via network views | ✓ gene-centric |
| Causal chain reconstruction (metab → rxn → gene → disease) | ✓ with posterior confidence | — | — | — | — |
| Counterfactual knockout (simulate removal, rescore) | ✓ | — | — | — | — |
| Calibrated couplings (learned on held-out triples) | ✓ | — | — | — | aggregated evidence scoring |
| Druggability + medicinal-chemistry handoff | ✓ | — | — | — | tractability links |
ORA and gene-centric knowledge tools excel at their individual jobs. GIZMO's contribution is the bridge; it rank diseases directly from a metabolite list with calibrated probability, reconstruct the biochemistry that justifies each ranking, and hand the top genes to a druggability stack without a manual hop through a different tool.
Upload your metabolite list, map to graph nodes, run analysis. The active graph is shared across every Forge user, so results are comparable across datasets.
Request Forge access →For biotechs with a compound panel and a disease area. We return a ranked target list with causal chains, counterfactual KO deltas, and druggability scores. Typically in 6–8 weeks from NDA signing.
Scope a project →Need a proprietary target set, phenotype ontology, or chemistry corpus integrated into the graph additively? We extend the active graph under contract, keeping public nodes intact.
Discuss a custom layer →GIZMO produces artifacts a medicinal chemist or clinical translator reads without reinterpretation: causal chains with posterior confidence, counterfactual deltas, ranked target lists with tractability annotations. It's the hand-off layer between the omics analysis and the next decision.
Benchmark suite and admin editor for custom-layer work ship with the platform. For Joey's essays and related work see insilijo.github.io.
code
github.com/insilijo/GIZMO
platform
github.com/insilijo/forge