We asked: STX-478 (mutant-selective allosteric PI3Ka) × PIK3CA
Dock the drug into the target’s binding pocket and score the pose.
Cross-species caveat. Docked against human 4JPS (PI3Kα / p110α). The canine PIK3CA ortholog (UniProt A0A5F4C2B1 / RefSeq XP_545208.2, taxid 9615) is 99.8% identical to human overall and 100% identical at every alpelisib/ATP-pocket residue — and a redock-verified canine receptor (AlphaFold AF-A0A5F4C2B1, alpelisib redock 0.62 Å) is now in the library — so the human structure is a justified cross-species proxy (the binding pocket is sequence-identical), not an unverified substitution.
Pre-registered tests, each locked to a kill-criterion before it ran. Any single one failing rules the whole idea out.
Survived every test the engine could run. Nothing here is proven — that's the point.
Readout. gnina docked STX-478 into PIK3CA: recommended pose -7.3 kcal/mol (CNN pose 0.70, CNN affinity 7.2); best Vina affinity -9.2 kcal/mol. Signal: supports.
Confidence so far: 32% — never proof. It survived the cheap lane(s) run so far; confidence stays low until deeper lanes run.
Every result is hashed so anyone can re-derive it. Clearing these checks is necessary, not sufficient — a human still holds the final call, and nothing is auto-promoted.
See the full experiment record.
The pre-registered kill-criterion, the raw readout, and the signed provenance you can verify yourself — plus every new verdict as it lands. Drop your email and we’ll open it up.
A mailing list for early access, not a login. We’ll email you to set up real sign-in once things are polished.