Intricacies in modifying term weights (e.g. FE_coeff_vdW)

[diogomart]

With the purpose of decomposing the ligand-grids (intermolecular) energy into the unweighted constituent terms (vdw, hb, elec, dsolv), autogrid 4.2.8 ran four times, once for each term, with the weight of one term set to 1.0 and the other three to 0.0. The score of a pose was calculated by multiplying each of the terms by the original AD4 weight, e.g. 0.1662 for vdW. This is the "weighted" approach.

Note: couldn't do this with 4.2.6 because it ignores the input weights. Setting FE_coeff_vdW 0.0 does not affect the output, even though FE_coeff_vdW banana raised a "can't convert to float" error.

In parallel, we ran autogrid 4.2.8 with defaults and simply got the total energy. This is the original approach.

The resulting intermolecular energies differed by about 0.3 kcal/mol for one particular pose. The HD maps were inspected because the unweighted contribution of one paricular HD atom differed by almost 10 kcal/mol depending on grid resolution (0.25 vs 0.375). This pose was straight from x-ray positions, an optimization would likely mitigate such strong dependence on grid resolution.

Differences in HD maps occur in the vicinity of OA and NA atoms, at distances inferior to 1.5 Angstorm. Maybe HD-OA and HD-NA pairs are not removed from vdW when FE_coeff_hb is zero.

Green: original HD map contoured at -0.1 kcal/mol
White: "weighted" HD map countoured at -0.1 kcal/mol. This is 0.1662 * vdW + 0.1209 * HB + 0.1322 * dsolv
Pink: weighted - original, countoured at +5 kcal/mol
Yellow: weighted - original but zeroed where original is positive, coutoured at +0.1 kcal/mol

Green and white agree where energies are negative, with the exception of a few yellow spots.

[atillack]

@diogomart Thank you Diogo! Good detective work :-)