Type a disease or pick one. Top-20 drug candidates ranked by the compressed model, with external corroboration for the 10 curated diseases.
How this is laid out. All 56 diseases have full corroboration: each candidate is cross-checked against ChEMBL drug indications and Open Targets disease associations. The 10 curated entries (marked ★) use strict disease-name matching; the other 46 are the highest-degree diseases in OGBL-biokg with looser matching that flags adjacent indications (e.g., a generic carcinoma drug counts as breast-cancer evidence). The looser matching trades precision for coverage — useful for browsing but read it as "any related indication" rather than "exact-disease indication".
#
Drug
PubChem CID
Score (mean ± std)
ChEMBL
OT
Corroborated
Important. Predictions corroborated by external databases are candidates worth a clinician's review — not validated treatments. Predictions not corroborated should not be assumed to be useless: they may lack database coverage, be research compounds, or be model artifacts. The model has no concept of dose, route, or patient state. Consult a clinician for any treatment decision.
Notes
Curated diseases: Type 2 diabetes, hypertension, hypercholesterolemia, coronary artery disease, asthma, Alzheimer's, multiple sclerosis, Parkinson's, breast cancer, schizophrenia. Each has full ChEMBL / Open Targets corroboration columns.
Score-only diseases: top-degree entries in the OGBL-biokg disease space, names resolved via MyDisease.info.
Disease degree (visible on hover): number of relations touching the disease in the train graph. Higher = the model has seen more about it.
Score is ComplEx-N3 mean over 3 random seeds. Std is the stability across seeds.