#4 Possible sign error in Smina's built-in AD4 hydrogen-bond coefficient (comparison with Gnina)

[Artur Duque Rossi]

Greetings,

While auditing scoring-function outputs for my molecular docking workflow, I identified a reproducible discrepancy between Smina's and Gnina's implementations of ad4_scoring.

Versions tested:

• Smina: Oct 15, 2019, based on AutoDock Vina 1.1.2
• Gnina: v1.3.2, commit f23dd2b

Both programs evaluate the same receptor and ligand PDBQT files using --score_only. Their reported unweighted energy terms agree to the printed precision, but the coefficient applied to non_dir_h_bond_lj differs.

Implementation	Hydrogen-bond coefficient
Smina built-in ad4_scoring	−0.097400
Gnina (recovered from energy decomposition)	+0.097439
AutoDock4.0 reference	+0.097400

I confirmed that the unweighted hydrogen-bond term becomes zero at large receptor–ligand separation and is negative for favorable hydrogen-bond interactions.

In one reproducible example, the unweighted hydrogen-bond term is −2.27907. The resulting scores are:

• Smina built-in AD4: −41.81483 kcal/mol
• Smina with only the hydrogen-bond coefficient changed to +0.0974: −42.25879 kcal/mol
• Gnina AD4: −42.25880 kcal/mol

Changing this single coefficient reproduces Gnina's result to within 0.00001 kcal/mol.

I also observed that the discrepancy varies systematically with the hydrogen-bond term across the complexes I tested, while the other AD4 term coefficients agree.

Could you clarify three questions?

1. Is the negative hydrogen-bond coefficient in Smina's built-in ad4_scoring intentional, or might it be a sign error?
2. Was this discrepancy already identified and corrected during Gnina's development?
3. If unintentional, would updating Smina's built-in AD4 configuration be appropriate?

AutoDock reference:
https://autodocksuite.scripps.edu/force-fields/

Thank you for your work on Smina and Gnina, and for taking the time to review this observation.

[David Koes]

This seems to be an honest-to-goodness bug. Surprisingly, I can't find the commit where it was fixed in gnina - it seems like it has been correct from the beginning. I will update smina.