Human acetylcholinesterase
The claim
Published docking studies on human acetylcholinesterase (AChE, PDB 4EY7) report dietary flavonoids as inhibitors with AutoDock Vina scores spanning roughly −8.6 to −12.6 kcal/mol. A few values stand out as unusually strong, notably rutin at −12.6 and epicatechin gallate at −12.2 kcal/mol.
The re-execution
We first confirmed the engine recovers the co-crystallised ligand (donepezil) in 4EY7 by self-docking: RMSD 0.27 Å, well below the 2 Å threshold. Twenty-two public phytochemicals were then re-executed with three fixed seeds (11, 29, 53), AutoDock Vina at exhaustiveness 64, GNINA deep-learning rescoring, PLIP interactions and a consensus score (ADS, 0–100). Seed-to-seed noise was low (median SD 0.03 kcal/mol).
This is a proof-of-concept pilot: only the literature anchor values supplied in our study brief are compared, and no published figure was invented.
Multi-angle reading
| Compound | Reported | Vina | GNINA | ADS | ± 2 | Interactions |
|---|---|---|---|---|---|---|
| Luteolin · CID 5280445 | — | -10.36 | 0.42 | 82.6 | — | 11 H-bonds, 6 hydrophobic |
| Quercetin · CID 5280343 | — | -10.18 | 0.26 | 77.9 | — | 12 H-bonds, 7 hydrophobic |
| Apigenin · CID 5280443 | — | -10.19 | 0.32 | 79.6 | — | 7 H-bonds, 7 hydrophobic |
| Kaempferol · CID 5280863 | — | -9.79 | 0.34 | 80.1 | — | 10 H-bonds, 7 hydrophobic |
| Rutin · CID 5280805 | -12.6 Vina (lit.) | -10.13 | 0.44 | 33.1 | ✗ | MW 610, Lipinski-fail, poor ligand efficiency |
| Epicatechin gallate · CID 107905 | -12.2 Vina (lit.) | -9.15 | 0.23 | 33.2 | ✗ | GNINA-contradicted; −12.2 only in a tight gorge box |
| Donepezil (reference) · CID 3152 | — | -11.33 | 0.69 | 76.2 | — | Co-crystal reference; self-docking 0.27 Å |
Vina in kcal/mol (lower = better) · GNINA CNN score 0–1 (higher) · ADS consensus /100 (higher). "± 2" = Vina deviation from the reported value within the 2.0 kcal/mol tolerance.
3D exploration
Fullscreen →Interactive companion: the target pocket in 3D, the real docking poses, and the whole pipeline (Vina, GNINA, drug-likeness, ML/DTI, consensus) explained step by step.
The finding
The mid-range flavonoid scores reproduce cleanly and stably (−9.8 to −10.4 kcal/mol). The eye-catching −12 values do not. Rutin's reported −12.6 is not reachable in any box tested here (best −10.1); epicatechin gallate's −12.2 appears only when the search box is tightly centred on the catalytic gorge: a whole-protein search gives −9.15, a 3 kcal/mol swing larger than the entire seed-noise budget.
Read by Vina alone, rutin (−10.1) and epicatechin gallate would tie the genuine flavonoids. The multi-angle consensus tells a different story: it demotes both to ≈ 33 / 100 (vs 76–83 for luteolin, kaempferol, apigenin) because of poor ligand efficiency, a Lipinski failure for rutin (MW 610), and a GNINA pose-quality contradiction for epicatechin gallate. The discrepancies track with under-specified search boxes, not with any error attributable to the authors.
Methods lesson
A single, very low Vina ΔG on a large or oversized phytochemical is more often an artefact of the search box or protocol than a real affinity. Reporting the box centre and size, plus a random seed, and reading a score alongside ligand efficiency and a second engine, is what turns a docking number into a re-executable, trustworthy claim.