Sage rc1 refit including dihedral_rmsd in optgeo target

Including the dihedral rmsd in the optgeo target’s objective function by making the dihedral_denom non-zero (a value of 10.0 degrees used here) improved the overall RMSD and TFD slightly when compared to Sage rc1. There is a slight degradation in ddE.

Open

Here are the benchmark plots on the Lim-Hahn optimization set:

Open

Open

Differences in parameters

Δ_Angles:

The maximum delta in the equilibrium angles is 2.95 degrees, and the maximum difference in k value is 37 units, here are those parameters

Percentage difference in values greater than 10%:

Δ_Bonds:

Not much difference, the maximum difference is in the bond length is 0.05 Angstrom, and the maximum difference in k value is 23 units, here are the parameters

All of the bond parameter changes are insignificant even in percentages,

Δ_Torsions:

Noticeable differences in torsion parameters, here is a list of parameters that show at least 0.5 kcal/mol difference in the k-values

[torsion_id, smirks_pattern, difference in k_values list_in_units_of_kcal_per_mol]

Percentage difference in any of the k-values is greater than 50%:

Inspecting some sulfonamide molecules

1. Here is one molecule that shows a good improvement with inclusion of dihedral_rmsd, wherein the hydrogens on the left NH2 are aligned properly in case of the refit.
   
   1a) 421_C6H7N2O5S2 in 1.2.0 training set
   assigned parameters: ['a10', 'a11', 'a18', 'a30', 'a31',
   'b14a', 'b5', 'b53', 'b54', 'b56', 'b84', 'b86',
   'c1', 'i1', 'i2a', 'n11', 'n14', 'n17', 'n20', 'n21', 'n7',
   't112', 't136', 't139', 't44']
   Green - QM, Cyan - Sage rc1

1b) 937_C15H20N2O3S in 1.2.0 training set

assigned parameters: ['a1', 'a10', 'a11', 'a17', 'a18', 'a19', 'a2', 'a30', 'a31',
'b1', 'b10', 'b2', 'b20', 'b3', 'b5', 'b53', 'b54', 'b56', 'b7', 'b83', 'b84', 'b86', 'b9',
'c1', 'i1', 'i2a', 'i3', 'n11', 'n14', 'n16', 'n17', 'n2', 'n20', 'n21', 'n3', 'n7',
't1', 't109', 't135', 't137', 't139', 't140', 't17', 't18', 't2', 't20', 't3', 't4', 't44', 't51', 't59', 't62', 't69a', 't70b']

Green - QM, Cyan - Sage rc1

Green - QM, Magenta - Sage refit w/ dihedral_rmsd

1c) 101_C4H10N2OS in 1.2.0 training set

assigned parameters: ['a1', 'a17', 'a18', 'a2', 'a23', 'a3', 'a30', 'a31', 'a4', 'a6',
'b1', 'b53', 'b54', 'b56', 'b7', 'b83', 'b86',
'c1', 'i2a', 'n11', 'n16', 'n17', 'n2', 'n20', 'n21', 'n3',
't109', 't134', 't135', 't137', 't139', 't140', 't145', 't15', 't16', 't54', 't55', 't56', 't57']

Green - QM, Cyan - Sage rc1

List of some systems (some are conformers) that have a high dihedral RMSD of > 40 degrees among sulfonamides from 1.2.0 targets that showed an improvement with the refit

2. Here is a molecule that shows a slight degradation:

1676_C7Cl2H5N3O2S3 in 1.2.0 training set
['a1', 'a10', 'a14', 'a19', 'a2', 'a20', 'a22', 'a30', 'a31', 'a35',
'b13', 'b2', 'b34', 'b4', 'b49', 'b53', 'b54', 'b56', 'b6', 'b69', 'b8', 'b83', 'b84', 'b86',
'c1', 'i1', 'i3', 'i4', 'n11', 'n14', 'n16', 'n17', 'n2', 'n20', 'n21', 'n24', 'n7',
't106', 't108', 't112', 't129', 't136', 't139', 't141', 't143', 't20', 't24', 't43', 't45', 't69a', 't70', 't77']

Green - QM, Cyan - Sage rc1

Green - QM, Magenta - Sage refit w/ dihedral_rmsd

The above molecule and another conformer of it showed a higher dihedral RMSD with the refit

Torsiondrives in Sage training set matching
[t26, t31, t64, t118, t135, t139, t154]

For the torsions listed above that show greater than 0.5 kcal/mol difference with the refit plotting the torsion profiles for the molecules in Sage training set. All residuals overlap in both cases, not much improvement/detriment in full MM torsion profiles or their RMSDs except in one or two cases below. This looks like a case of multiple minima, where the optimized geometries are improved with the inclusion of dihedral rmsd in optgeo target’s objective function but with a change in the torsion parameter values no significant change observed.