High-Throughput Prediction of Single-Point Energies for Charged Polymer Monomers through Feature-Engineered Machine...

Area of Science:

• Computational Chemistry
• Materials Science
• Polymer Science

Background:

• Accurate prediction of single-point energies is crucial for understanding molecular stability and electronic structure, especially for polar and charged polymeric monomers.
• Density Functional Theory (DFT) offers reliable energy estimates but is computationally expensive for large, diverse chemical systems.

Purpose of the Study:

• To develop a DFT-assisted machine learning framework for predicting single-point energies of charged polymer monomers.
• To create a robust model capable of handling large chemical spaces and varying dielectric environments.

Main Methods:

• A curated dataset of charged polymer monomers was generated via PubChem screening.
• Cheminformatics and DFT-derived electronic descriptors were combined and filtered to identify optimal feature subsets.
• Gaussian process regression was employed as the primary supervised learning model.

Main Results:

• The Gaussian process regression model demonstrated excellent agreement with DFT calculations on both training and test sets.
• The framework maintained high accuracy across multiple dielectric environments.
• Atomic contribution analysis and Shapley additive explanations provided structural insights into energy relationships.

Conclusions:

• The developed DFT-assisted machine learning framework offers a computationally efficient and accurate alternative to DFT for predicting charged polymer monomer energies.
• This approach enables rapid exploration of extended chemical spaces and aids in understanding structure-energy relationships.