Accuracy evaluation and addition of improved dihedral parameters for the MMFF94s

Abstract

The Platinum dataset of protein-bound ligand conformations was used to benchmark the ability of the MMFF94s force field to generate bioactive conformations by minimization of randomly generated conformers. Torsion angle parameters that generally caused wrong geometries were reparameterized by conducting dihedral scans using ab initio calculations at the MP2 level. This reparameterization resulted in a systematic improvement of generated conformations.

Keywords: Conformer generation; Force field parameterization; MMFF94s.

Fig. 1

Example representations of torsion fragments. Example 1 refers to the dihedral angle of the rotation around a C–C bond, whereas C₁ is part of a delocalized system, neighboring two C-atoms with two neighbors each (ⁿ²C). C₂ is a non-aromatic C (^!aC) bound to a nitrogen atom. In example 2, we again refer to the rotation around a C₁–C₂ bond. C₁ is a non-aromatic carbon, bound to a non-aromatic carbon with less than 4 neighboring atoms (C^!a,n<4; sp² or sp hybridized). C₂ is an amide carbonyl carbon. Example 3 is a combination of Examples 1 and 2, where C₁ is part of a delocalized system, neighboring two Cⁿ² atoms, and C₂ is an amide carbonyl carbon

Fig. 2

Torsional fragments for which the MMFF94s-minimized structures exhibited systematic deviation of the central dihedral angle compared to the experimental structures

Fig. 3

Torsional energy profiles for various substructures with problematic torsion angles for the MMFF94s. The four atoms defining the scanned torsion angles are indicated by asterisks. Conformational energies are given in kJ/mol. These structures are not present as ligands in the platinum dataset. Torsion histograms from the CSD with relative frequencies were added for several compounds

Fig. 4

Plot of RMSDs between the closest low-energy conformer and the bioactive conformation for the original MMFF94s against the MMFF94s with the new parameters (MMFF94s_new) for 2833 structures from the Platinum Diverse Dataset. Data points where the unsigned difference between the two RMSD values is less than 0.1 are colored orange, data points where the MMFF94s_new performs better are shown in blue, whereas the grey data points indicate a better performance by the original MMFF94s