Selective recognition between aromatics and aliphatics by cage-shaped borates supported by a machine learning...

Area of Science:

• Organic Chemistry
• Computational Chemistry
• Catalysis

Background:

• Selective recognition of hydrocarbon moieties is a significant challenge in chemistry.
• Existing Lewis acid catalysts lack a generalizable strategy for precise molecular recognition.

Purpose of the Study:

• To develop a rational catalyst design strategy for selective hydrocarbon recognition using machine learning.
• To enhance the selectivity of Lewis acid catalysts beyond current limitations.

Main Methods:

• Employed an ensemble machine learning algorithm (Random Forest, Ada Boost, XG Boost) to predict catalyst chemoselectivity.
• Utilized 7963 explanatory variables from model hetero-Diels-Alder reactions for catalyst feature extraction.
• Applied the SHapley Additive exPlanations (SHAP) method to interpret model predictions.

Main Results:

• Achieved highly selective Lewis acid catalysts for hydrocarbon moiety recognition, outperforming previous methods.
• Experimental validation confirmed the machine learning model's predictions for catalyst performance.
• Identified catalyst polarizability and three-dimensional size as key factors driving selectivity.

Conclusions:

• The integration of machine learning with non-covalent interactions provides a powerful framework for rational catalyst design.
• The developed catalysts demonstrate enzyme-like selective recognition capabilities.
• Established design guidelines for future Lewis acid catalysts, emphasizing dispersion forces and specific structural attributes.