Name a disease gene or a condition. A model proposes drug targets, a second model reviews each one,
and every target comes back with drugs you can already order, from 10,794 approved and clinical-stage compounds.
Audit direction + evidencea second, different model
Match compoundsclinical library → ChEMBL 37
Flag + rankPAINS, selectivity, mechanism direction
Research use only. These computational results are not medical advice, clinical guidance, or a therapeutic recommendation.
Targets identified by LLM reasoning about disease biology. Compounds matched from our clinical library (predicted activity) or ChEMBL (measured activity). Click a header to expand/collapse.
Target hypotheses
One model proposes targets; a second, different model audits each one, drops directions that would
deepen the lesion, and rewrites the rationale to fit the evidence.
Compounds
10,794 compounds ranked by a 756-target pharmacology model, else measured ChEMBL 37 activity.
Cut-off pIC50 5.0, 10 per target. "Approved" = listed in Drugs@FDA or the FDA Orange Book,
never a trial phase.
Annotations
BBB class and PAINS, promiscuity and weak-potency flags per compound. Active-trial badges from a
monthly ClinicalTrials.gov index. Literature via
PubTator3,
sourcing via PubChem.
Reading a result
Nothing is filtered for being uncertain — weak hypotheses are graded, not removed.
Methods · how to cite.
Models update in place; record the date.
These computational results are strictly for laboratory research purposes. They do not constitute
medical advice, clinical guidance, or therapeutic recommendations.