Re-running default docking with a new seed can change which compounds make the shortlist
- A single-target audit re-ran a default AutoDock Vina pipeline on LIT-PCBA MTORC1 with a small ligand panel, changing only the random seed, search box position or protomer.
- Seed and box shifts changed several members of the top-ranked set, while protomer changes did not; a few ligands caused all the flips.
- A majority vote across several seeds cost several times the compute and still disagreed with every single run.
- The authors stress this measures repeatability only, not correctness.
💡 Points Based on the abstract
That a replicate vote still disagrees with every single run says the shortlist is partly seed luck, and no run, voted or not, is a stable reference. A useful warning for anyone treating one default Vina run as the shortlist. The protomer result is reassuring, but it is one target and a small panel, and repeatability is not correctness. A sensible habit: run replicates and send the few unstable ligands to manual review.