Independent validation campaigns
Summary of 12 docking campaigns
# | Campaign | n | Result |
1 | Initial 1,397-drug screen at pocket #1 (exhaust=1) | 1,397 | top 50 ranked by selectivity |
2 | Robust check top 10 (exhaust=8) | 10 + 2 ctrl | only phenytoin survives |
3 | Extended top-30 robust expansion | 13 | +1 borderline hit (DEXLANSOPRAZOLE) |
4 | Phenytoin × 5 pockets (cross-cavity) | 5 | pockets #1, #7, #19 selective; #3, #15 not |
5 | Pocket #19 multi-ligand test | 4 | apparent strong signals |
6 | 19 AEDs at pocket #19 | 19 | 14/19 robust (suspicious) |
7 | Top 50 retest at pocket #19 | 14 | 14/14 robust (very suspicious) |
8 | Random control at pocket #19 + pocket #1 | 7 × 2 | DECISIVE — pocket #19 permissive |
9 | 19 AEDs at pocket #1 (H1 validation) | 19 | only 2 robust (phenytoin + clonazepam) |
10 | Pharmacophore extension (benzo + hydantoin) | 11 | +1 robust hit (OXAZEPAM) |
11 | Pharmacophore stress test (extended) | 25 | 0/25 robust → confirms specificity |
12 | Publication-grade re-docking exhaust=16/32 | 4 | identical Δ → docking converged |
Why this layered approach matters
A single robust hit could be a stroke of luck. The combined evidence:
- Within-class control: 17/19 AEDs do NOT bind selectively. Phenytoin is not generically selective for "all anti-epileptics" — it's specifically selective.
- Random control: 0/7 random FDA drugs bind at pocket #1. The pocket discriminates, doesn't take any druglike compound.
- Chemical-analogue control: 0/25 chemically-close compounds bind selectively. The pharmacophore is narrow.
- Convergence control: Δ values identical at exhaust=8, 16, and 32. The docking has converged; we're not still sampling.
- Cross-cavity control: pocket #19 binds nearly anything → can be excluded as a hypothesis vector.
- Pocket #19 reframe: documented openly, even though it was an apparent positive result.
A complete data table is on OSF
Every Δ value cited above traces back to one of these CSV files: aed_pocket1.csv (19 AEDs at pocket #1), aed_pocket19.csv (19 AEDs at pocket #19), aed_pocket7.csv (cross-pocket validation), random_pocket1.csv and random_pocket19.csv (random controls), pharmacophore_pocket1.csv (11 chemically-related), extended_pocket1.csv (25 chemically-related stress test), top30_robust_e8.csv, top20_pocket19.csv (top 50 retest at pocket #19), pocket1_pub_grade.csv (exhaust=16/32 convergence). Full file listing on page 13.