Pocket #19 — a cautionary tale about apparent super-targets

When phenytoin was docked against all 5 detected pockets in the R203W complex, one cavity stood out:

Pocket #19 produces a signal more than twice as strong as pocket #1, and is located directly at the W203 mutation site (residue 104 ESMFold + furin tail contacts).

We tested more compounds at pocket #19:

We extended to all 19 FDA antiepileptics: 14/19 (74%) robustly selective, including drugs that are NOT robust at pocket #1 (carbamazepine, lamotrigine, lacosamide, ...). We re-tested the top 50 hits from the initial screen at pocket #19: 14/14 (100%) robust.

This looked like a discovery. Then we ran a critical control.

We tested 7 randomly-selected FDA drugs (not in any prior heuristic) at both pockets:

Pocket #1: 0/7 random robust → discriminates.

Pocket #19: 7/7 random robust → permissive.

Pocket #19 is real biology (a cryptic cavity exposed by the R→W mutation). But it binds nearly anything druglike with a systematic ~−1 kcal/mol shift. It is NOT a useful discriminating target — it cannot distinguish phenytoin from a random anti-inflammatory.

Pocket #1 is a smaller signal (−0.58 kcal/mol) but it discriminates among chemically diverse drugs. Phenytoin selective (−0.58), carbamazepine not selective (+0.09), random drugs not selective (Δ ≈ 0). This is the real result.

This is a textbook case of a finding that looks spectacular at first, then collapses under proper controls. Without the random control, we might have published pocket #19 as a major discovery. With it, we correctly downgrade the pocket #19 therapeutic target claim from high to low reliability (30% in our final scoring). This reframe is documented openly. We believe transparency about apparent findings that didn't survive controls is essential for trustworthy science.