openforcefield/openff-forcefields: Version 1.3.1-alpha1 "Parsley" Update candidate

Short summary

The new force field files in this release are adapted from the openff-1.3.0 release, except that two angle parameters have been reverted to their original values from smirnoff99Frosst 1.1.0.

The ability of the Parsley line of force fields ability to reproduce the tetrahedral geometry of certain sulphonamides (especially primary sulphonamides) has progressively and subtly degraded between releases, cumulating in OpenFF 1.3.0 yielding a O-S-N valence angle of < 90 deg for certain molecules while angle of ~109.5 would be more expected.

Simulating CC1=CC=C(C=C1)S(=O)(=O)N in vacuum using OpenFF 1.3.0, for example, shows an average O-S-N valence angle of 76.26 (+/- 0.21)

This issue has been tracked to the a30 ([*:1]~[#16X4:2]~[*:3]) and a31 ([*:1]-[#16X4,#16X3+0:2]-[*:3]) valence angle parameters, whereby the equilibrium angle has monotonically reduced while simultaneously the force constant has increased:

 ---------------------------------------------------------- 
|     | SMIRNOFF99Frosst |  Parsley 1.2.0 |  Parsley 1.3.0 |
|     |------------------|----------------|----------------|
|     | k     | theta    | k     | theta  | k     | theta  |
|-----|-------|----------|-------|--------|----------------|
| a30 | 140.0 | 109.5    | 201.1 | 104.2  | 220.7 |  99.1  |
| a31 | 120.0 | 109.5    | 152.3 |  96.3  | 188.4 |  89.8  |
 ---------------------------------------------------------- 
(k in kcal / mol, theta in deg)

This has seemingly in turn resulted in the degradation of protein-ligand binding free energies for certain molecules which contain sulphonamide moieties (primarily primary sulphonamide groups). This potentially also affects related hypervalent substituents (e.g. hypervalent nitrogens) although currently only issues with sulphonamides has been observed.

We believe this issue arose due to the valence angle being trained to both sulphur containing rings in addition to molecules where the sulphur would be expected to be tetrahedral, whereby ring containing sulphur molecules would likely drive the angles to be smaller than 109.5 degrees. We are now actively working to validate this hypothesis and come up with a sustainable resolution for future releases.

For now given the potential severity of the issue, it has been decided that we will make an alpha patch release of OpenFF 1.3.0, numbered OpenFF 1.3.1-alpha.1, which has the values of a30 and a31 reset to their smirnoff99Frosst 1.1.0 values.

Provided that benchmarks show this issue as resolved OpenFF 1.3.1-alpha.1 will become a full release -> OpenFF 1.3.1.