Re: [Rdkit-discuss] UFF and MMFF94 strongly disagree on the internal energy of one conformer

[Geoffrey H. <geo...@gm...>]

>
> I am a little bit annoyed by the fact that the UFF energy
> is not even negative and that the two FFs disagree by so much.
>

I would not consider this to be a big problem, IMHO.

Do the two FFs compute something different?
>

Yes, very much so. UFF, for example does not include electrostatic
non-bonded terms, and MMFF94 does. Additionally, MMFF94 includes a specific
term for hydrogen bonding interactions.

I consider force field energies to only be relevant in relative terms. That
is, taking the relative difference of energy of relative conformers using a
force field is more useful than the absolute value. (And particularly
comparing between numbers from UFF vs. MMFF94 vs. GAFF, etc.)

Even then, while force fields are useful for “quick cleanup” as my group
showed (https://doi.org/10.1002/qua.26381) the relative energy rankings of
UFF and MMFF94 are still crude relative to ML or quantum methods.

In other words, the question is “does optimizing a conformer with UFF or
MMFF94 help?”

Yes. In general, conformers generated using RDKit (or Open Babel .. or most
methods I’ve seen) have lower RMSD when optimized by a force field.

Hope that helps,
-Geoff

---
Prof. Geoffrey Hutchison
Department of Chemistry
University of Pittsburgh
tel: (412) 648-0492
email: ge...@pi...
twitter: @ghutchis
web: https://hutchisonlab.org/