Drug discovery costs $2.6B per approved molecule and takes 12–15 years. Drug repurposing — finding new clinical uses for existing compounds — is the most efficient acceleration path. The bottleneck is hypothesis generation: existing ML tools require large training sets of known drug-disease pairs, produce black-box scores with no mechanistic explanation, and cannot generalize to drugs or diseases not seen during training.
EasyAtom is a 16-layer algebraic causal reasoning engine that generates drug repurposing hypotheses from a 6M-triple biomedical knowledge graph — without ever seeing a single drug-disease "treats" edge as input. Every output comes with a fully auditable causal chain: drug → gene → pathway → disease, reproducible on standard hardware.
| Method | Recall@10 | Trains on drug-disease data? |
|---|---|---|
| Random baseline | 4.7% | No |
| EasyAtom v4.3 | 21–28% | No — fully zero-shot |
| TransE (transductive) | ~31% | Yes — trained on 80% of pairs |
| RotatE / DRKG | 38–42% | Yes — trained on 80% of pairs |
| CompGCN / GNNs | 45–65% | Yes — trained on 80% of pairs |
Experimental validation partner. We seek a pharma, biotech, or academic lab willing to test 1–3 top predictions in vitro (cell assay or binding assay). We provide the full causal chain and ranked rationale. A confirmed or refuted result is publication-worthy regardless of outcome. No equity required — open to sponsored research agreement, MTA, or co-authorship arrangement.