Developing a Machine Learning Model for Hydrogen Bond Acceptance Based on Natural Bond Orbital Descriptors.

Area of Science:

• Computational Chemistry
• Machine Learning in Chemistry
• Quantum Chemistry

Background:

• Hydrogen bonding plays a crucial role in molecular interactions and chemical processes.
• Predicting hydrogen bond acceptance strength is vital for understanding molecular behavior.
• Traditional methods often require complex descriptors and extensive quantum chemical calculations.

Purpose of the Study:

• To evaluate the predictive capability of electronic descriptors from Natural Bond Orbital (NBO) analysis for hydrogen bond acceptance.
• To develop accurate and generalizable machine learning models using NBO-derived features.
• To establish a novel approach for predicting pKBHX values using a simplified descriptor set.

Main Methods:

• Geometry optimization of 979 hydrogen bond complexes using GFN2-xTB.
• Density Functional Theory (DFT) single-point calculations followed by NBO analysis.
• Extraction of orbital stabilization energies (E(2)) as key machine learning descriptors.
• Training and evaluation of seven machine learning models (KNN, Decision Tree, SVM, RF, MLP, XGBoost, CatBoost).

Main Results:

• NBO-derived orbital stabilization energies (E(2)) proved effective as standalone machine learning descriptors.
• Achieved high predictive performance with prediction errors below 0.4 kcal mol⁻¹.
• Outperformed previous studies utilizing more complex and heterogeneous descriptor sets.

Conclusions:

• NBO-based electronic descriptors are highly effective for building accurate machine learning models.
• This approach offers a physically meaningful and generalizable method for predicting hydrogen bond acceptance.
• The study demonstrates the utility of NBO analysis in advancing predictive modeling for molecular properties.