Accurate Prediction of Three-Body Intermolecular Interactions via Electron Deformation Density-Based Machine Learning

Area of Science:

• Computational Chemistry
• Machine Learning in Chemistry
• Quantum Chemistry

Background:

• Accurate prediction of molecular interactions is crucial in chemistry.
• Existing methods for calculating three-body interactions can be computationally expensive.
• Machine learning (ML) offers a promising avenue for accelerating these calculations.

Purpose of the Study:

• To extend the EDDIE-ML algorithm for predicting three-body interactions in trimers.
• To develop a ML model capable of predicting three-body interaction energies with high accuracy.
• To introduce a novel, larger dataset for benchmarking trimer interactions.

Main Methods:

• Utilized an electron deformation density-based descriptor within a Gaussian process regression (GPR) model.
• Employed a sequential learning process for efficient training data selection.
• Introduced a hybrid kernel function combining average and individual atomic environments.

Main Results:

• The GPR model achieved high accuracy, predicting three-body interaction energies within 0.2 kcal/mol of reference values.
• The model successfully predicted both three-body contributions and total trimer interaction energies.
• A new dataset of 509 protein-ligand structures was introduced, featuring larger and charged molecular interactions.

Conclusions:

• The developed ML model provides an efficient and accurate method for predicting three-body interactions.
• The approach significantly reduces computational cost compared to traditional DFT and wavefunction methods.
• This work demonstrates the utility of ML for accelerating complex molecular interaction calculations, especially for larger systems.