concept · intermediate · partly written
PrimeKG
PrimeKG is a drug-disease-gene knowledge graph assembled from 20 curated resources, with 129,375 nodes and 4,050,249 relations.
Also called: Precision Medicine Knowledge Graph
Short answer
PrimeKG (Chandak, Huang, Zitnik, 2023) has 129,375 nodes of 10 types and 4,050,249 relations of 30 types from 20 resources, built from curated databases without text mining. Zitnik is the last author. The paper reports no link-prediction benchmark, and no per-edge evidence is described. 1
Reported, not yet replicated
What its authors checked more
The data paper's checks are small: 90.8% of 9,348 Orphanet rare diseases are present; relevant drugs sit nearer to their disease than random in 8 of 10 testable pairs (an independent read of the table counted 7 of 10 at adjusted P below 0.05, and the paper wavers between 10 and 11 pairs); of 40 FDA approvals after the cut-off, 11 were repurposed drugs and one pair already had an indication edge. Its claim of a one-to-two orders of magnitude disease-coverage advantage is the authors' assertion. The graph is frozen at about June 2021. 12
Contested
How names become identifiers more
Diseases are mapped to MONDO and grouped (22,236 terms into 17,080) by string match, ClinicalBERT similarity of at least 0.98 and manual approval; no accuracy figure appears in the text read. The authors found MONDO listing 37 autism concepts, UMLS 192 and Orphanet 6. 1
Reported, not yet replicated
Use by TxGNN, and its caveats deep
TxGNN (Huang, Zitnik and colleagues) was trained on a PrimeKG derivative. Its headline gains are the authors' own; the preprint body gives 19.0% and 23.9% while its abstract gives 49.2% and 35.1%, which this atlas could not reconcile. 2
Reported, not yet replicated
Still open here
- Do PrimeKG-based link-prediction scores survive a degree-only baseline and a by-disease split run by someone outside the builders? No degree baseline for PrimeKG was found.
- Does the TxGNN model run on PrimeKG itself? One report states it; a 123,527-node derivative is also described; treat as probable, not confirmed.
Not yet written: Plain answer, Why it matters, Equation, Real numbers.
Linked from
- Biomedical knowledge graphs: shapes, edges and what has worked
- How a link-prediction benchmark is built and how it leaks
- The QIAGEN Knowledge Base (behind Ingenuity Pathway Analysis)
- Is orthology or evolutionary information an edge type anywhere, and does it help prediction?
- Has any top prediction from a public biomedical graph been tested prospectively?
- What fraction of a graph benchmark score survives a degree-only baseline and a by-disease split?
Sources
- Chandak P, Huang K, Zitnik M. Building a knowledge graph to enable precision medicine. Scientific Data 10:67 (2023). paper; opened 2026-11-18 all records citing this
- Huang K, et al., Zitnik M. Zero-shot drug repurposing with geometric deep learning and clinician centered design (TxGNN, medRxiv version 4). medRxiv (PMC11326339) (2024). preprint, not peer reviewed; opened 2026-11-18 all records citing this
Where it sits on the map
History of this record (6)
This record has been changed 6 times, on the dates below. The note kept against each change is written for the people who maintain the atlas, in their own shorthand, so it is not shown here; it is in the record itself.
- Status
- partly written
- Last updated
- 2026-11-19
- Record
- content/concepts/primekg.json