Deep Neural Network Model to Predict the Electrostatic Parameters in the Polarizable Classical Drude Oscillator Force...

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Drug Discovery

Background:

• The Drude polarizable force field (FF) models electronic polarization using auxiliary Drude particles, differing from additive FFs with fixed charges.
• Current Drude FF parameterization is challenging for novel small druglike molecules, requiring partial charges, atomic polarizabilities, and Thole scaling factors.

Purpose of the Study:

• To develop deep neural network (DNN) models for predicting Drude FF electrostatic parameters for small druglike molecules.
• To enable rapid estimation of parameters essential for accurate molecular simulations.

Main Methods:

• Trained DNN models on quantum mechanical (QM) data for partial charges, atomic polarizabilities, and Thole scale factors.
• Utilized molecular connectivity and atom types as feature vectors to capture local and nonlocal effects.
• Developed novel methods for determining restrained electrostatic potential (RESP) charges and Thole scale factors.

Main Results:

• DNN models accurately predict molecular dipole moments and polarizabilities for FDA-approved drugs.
• Validation against high-level QM calculations (MP2) demonstrates high precision.
• The models effectively capture local and nonlocal electronic effects.

Conclusions:

• DNN models provide a rapid and accurate method for estimating Drude electrostatic parameters.
• This facilitates the broader application of the Drude polarizable force field to diverse molecular systems.
• Accelerates the use of advanced molecular simulation techniques in drug discovery and development.