Increasing the Accuracy and Robustness of the CHARMM General Force Field with an Expanded Training Set

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Drug Discovery

Background:

• Empirical force fields (FFs) like CGenFF are crucial for molecular simulations, requiring accurate parameterization for diverse organic molecules.
• Current CGenFF parameter assignment relies on interpolation and analogy, with accuracy dependent on the training set's breadth.
• Expanding the training set is key to improving the predictive power of CGenFF for novel molecular structures.

Purpose of the Study:

• To extend the CGenFF training set with 1390 new molecules featuring novel connectivities.
• To develop and validate a new version, CGenFF v5.0, incorporating updated parameters and charges.
• To assess the performance of CGenFF v5.0 against experimental and quantum mechanical (QM) data for various molecular properties.

Main Methods:

• Quantum mechanical (QM) calculations were performed for optimized geometries, potential energy scans, dipole moments, and electrostatic potentials.
• A new training set of 1390 molecules was curated to cover previously unrepresented chemical connectivities.
• CGenFF v5.0 was trained using QM data and validated against experimental and QM benchmarks for drug-like molecules.

Main Results:

• CGenFF v5.0 demonstrates significant improvements in QM intramolecular geometries, vibrations, dihedral scans, dipole moments, and water interactions.
• Minor enhancements were observed in pure solvent properties compared to CGenFF v2.5.1, highlighting robust Lennard-Jones parameters.
• The new version shows better performance in modeling intramolecular strain energies and noncovalent interactions.

Conclusions:

• CGenFF v5.0 offers enhanced accuracy for simulating organic molecules, particularly drug-like compounds.
• The expanded training set and refined parameterization lead to more reliable predictions of molecular behavior.
• This advancement is expected to improve the accuracy of molecular modeling in computational chemistry and drug design.