This is an open computational-methods exercise. Corrections, replications, and methodological critiques are genuinely welcome — the point is to do structure-based triage in public and get the caveats right.
- Replicate the docking. Re-run
scripts/reproduce.shand report whether the selectivity direction holds (hydroxamates anti-selective, sulfamides ENPP1-preferring). Exact scores will drift ≈0.1–0.3 kcal/mol run to run; the ranking is the claim, not the digits. - Stress-test the protocol. Different box size, exhaustiveness, scoring function (e.g. Vina vs smina default), or an MD-relaxed pose. Does the flip survive?
- Improve the metal treatment. The biggest known weakness is that smina models Zn coordination poorly. A QM/MM or explicit-metal-restraint pass on the top compounds would materially strengthen (or overturn) the conclusion.
- Add compounds. More sulfamides / non-chelating chemotypes to test whether the margin-vs-score pattern generalizes.
This repo makes no therapeutic claim and contains no wet-lab data. Any PR
that presents docking ΔΔG as an IC50 ratio, or the nominee as a validated
candidate, will be asked to reword. The credibility of the piece rests on the
limitations being stated plainly — see the DISCLAIMER in README.md.
Open an issue with: what you ran, the environment (OS, smina/obabel/RDKit versions), and the output. For scientific disagreements, cite the structure or paper you're arguing from.