Accurate prediction of polarised high order electrostatic interactions for hydrogen bonded complexes using the...

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Machine Learning Applications

Background:

• Intermolecular interactions, like hydrogen bonds, are fundamentally electrostatic.
• Accurate electrostatic treatment is essential for robust force field methodologies.
• Existing methods may not fully capture the nuances of these interactions.

Purpose of the Study:

• To develop and validate a novel method for accurately reproducing electrostatic intermolecular interactions.
• To assess the performance of this method across different theoretical levels and molecular systems.
• To improve the accuracy of force fields by better modeling electrostatic contributions.

Main Methods:

• Utilized atomic multipole moments up to the hexadecupole level.
• Employed the machine learning method kriging to map multipole moments to nuclear coordinates.
• Developed models at three quantum chemical levels: HF/6-31G(**), B3LYP/aug-cc-pVDZ, and M06-2X/aug-cc-pVDZ.

Main Results:

• Achieved >90% prediction accuracy within 1 kJ/mol for small van der Waals complexes across all theory levels.
• Showed 60-70% accuracy for larger base pair complexes.
• Models based on B3LYP and M06-2X levels generally outperformed the HF level models.

Conclusions:

• The kriging-based approach effectively reproduces electrostatic interactions for van der Waals complexes.
• This method offers a significant improvement for force field development, particularly for hydrogen bonding.
• The accuracy achieved demonstrates the potential of machine learning in computational chemistry for electrostatic modeling.