OmiKG is a domain-specific knowledge graph that traces chronic disease back to its root causes — mapping the directed causal web from lifestyle and environment all the way down to clinical phenotype.
Existing biomedical ontologies (SNOMED CT, ICD, Gene Ontology) mostly describe "is-a" relationships — great for classification, blind to cause. Chronic disease is a network disorder. OmiKG models the directed, multi-factor causal chains needed to reason from a symptom back to its true origin.
OmiKG models directed causal chains (A→B→C), not just taxonomies — so it reasons about why a symptom exists, not merely what to call it.
Walk backwards from clinical phenotype to dysfunction to upstream trigger — true root-cause diagnosis, not surface management.
Each causal link carries a reference code back to its source passage — a transparent map that black-box models cannot reproduce.
N-ary causal modeling captures how diet, deficiency or toxins independently converge on the same disease — beyond simple binary triples.
A PRISMA-guided pipeline unifies thousands of PubMed passages before extracting knowledge — cutting noise and hallucinated relations.
Built on the Functional Medicine Matrix and P4 medicine — predictive, preventive, personalized and participatory by design.
Four layers and 25 categories, connected by 14 standardized relation types. The graph flows from upstream drivers down to clinical outcomes — and can be read in both directions.
Lifestyle & environmental drivers: chronic stress, diet, toxins.
Core biological disturbances: HPA-axis dysregulation, chronic inflammation, oxidative stress.
Physiological systems: endocrine, digestive, immune.
Clinical manifestations: insulin resistance, type 2 diabetes.
In a blinded expert evaluation, OmiKG identified the molecular mechanisms behind chronic disease far more completely than a leading LLM or a retrieval baseline.
OmiKG is a long-term research program. Every time we ship a new version of the OmiKG Core, we announce it here.
First public release of the four-layer ontology and knowledge graph: 1,354 nodes and 2,492 validated triples, built from a PRISMA-guided synthesis of PubMed literature.
OmiKG identifies 88% of expert-specified molecular mechanisms — including NLRP3/IL-1β, SREBP-1c and BCAA/mTOR pathways missed by other systems.
Next versions move beyond Type 2 Diabetes to Metabolic Syndrome and PCOS, and will quantify causal strength — not just confidence that a link exists.
OmiKG is being developed by OmiGroup as long-term research infrastructure for detecting disease before symptoms appear. Partner with us, or follow the research.
