2024-11-06 FF fitting meeting

Phosphate issue update

AMI

• (CB, pre-presentation: group meeting at OpenEye now clashes with this meeting so it’ll be hard for me to make it regularly)

• CB: is capital K the force constant?

  • AMI: yes, in kcal/mol/rad2

• CB: is phosphonate O being pulled into -OH? An electrostatic attraction that overwhelms the angle?

  • AMI: yes, they basically all look like that

  • CB: so maybe in the MM the issue is that the electrostatics are dominating, which is not present in the QM wavefunction

  • …

  • CB: So we’re titrating between force constants to prevent the valence angle from being overwhelmed?

    • AMI: yes

  • JR: how well do charges reproduce dipole moment?

    • JR: I don’t like AM1 (and -BCC), especially for P and S. First question is if it gets the dipole moment right

    • CB: if we can understand what’s going on better, maybe it will help us figure out a solution. I’m wondering if the defect has more to do with atom-centered charges and the strong attraction between O and H. For these particular molecules it just happens to be strong enough that it affects the angle unless we set the k artificially high. I see the H-O-P angle is not distorting, just the O-P-O.

    • JR: IMO the charge density here needs to be described using more than just charge. openFF is discussing off-centered charges, which seems reasonable to me, but wondering if we should look at “that” [off-centered charge / multipoles] as final solution

    • CB: if we think we’re going down a singularity caused by the H-O-P-O pattern, I would look at making a bond angle that’s #1-#8:1-#15:2-#8:3. Make an OPO angle specific to HOPO. Replaces good physics with more specific smarts pattern. Leave other parameter to be more general.

    • AMI: think this is a good idea. This is the framework we’re more aligned with.

  • JR: would this affect how the molecule H-bonds with others?

    • AMI: presumably. Typically 1 conformer out of ~5 will show this behaviour. Seemingly random, depends on how the O optimizes. Not universal issue for these molecules.

• CB: if you freeze a40 at the MSM value, does that make it low or high?

  • AMI: high. It was 180

  • CB: Concerned it may make the general angle too stiff. Suggest freezing H-O-P-O parameter, but let OPO be data-driven.

    • AMI: need to check if we have enough data

    • CB: the angle is for a phosphoric acid. pKa is so high for phosphoric acid it’s rare to work with, usually would work with an ionised version. I’m conscious that we don’t want to let the rare case pull us away from the more common P angles.

    • AMI: will look into it

• JR: so why are we looking at this neutral version and not the anion?

  • AMI: this is from the industry benchmarking set.

  • CB: a medicinal chemist makes and registers a phosphoric acid, but it ionises once you put it into a buffer or enzyme assay. It wouldn’t occur in biological milieu. Same issue with sulfonic acids.

•  

• CB: want to ask about fitting philosophy. OpenEye is looking into bespoke fitting. I hate it, it admits defeat in the general FF. OpenFF is working to improve a general FF so we need bespoke fitting less. Does OpenFF have a larger strategy for recommending when to use a bespoke FF? Bespoke fitters still want to generalize their parameters, and AMI here is splitting out parameters for more specificity

  • BS: came up recently that we could have a confidence metric for a particular compound that looks at whether it’s represented in the training set. Is a similar tool what you’re after?

  • CB: that’s an end product, at the end of this discussion.