Machine Learning and Quantum Calculation for Predicting Yield in Cu-Catalyzed P-H Reactions

Area of Science:

• Computational Chemistry
• Catalysis
• Machine Learning

Background:

• Copper-catalyzed P-H insertion reactions are crucial in synthesis.
• Predicting reaction transition states and yields is challenging.
• Accurate prediction can optimize reaction conditions and discovery.

Purpose of the Study:

• To develop predictive models for copper-catalyzed P-H insertion reactions.
• To utilize machine learning and quantum chemistry for yield prediction.
• To understand transition state characteristics influencing reaction outcomes.

Main Methods:

• Density Functional Theory (DFT) for transition state determination.
• Machine learning algorithms, including Support Vector Machine (SVM), for yield prediction.
• Analysis of 16 descriptors derived from quantum chemical calculations.

Main Results:

• Support Vector Machine (SVM) achieved 97% prediction accuracy for product yield.
• Over 80% correlation was observed using Leave-One-Out Cross-Validation (LOOCV).
• Sensitivity analysis identified key descriptors influencing reaction outcomes.

Conclusions:

• Machine learning models, particularly SVM, effectively predict yields of copper-catalyzed reactions.
• The study provides insights into reaction mechanisms and transition state properties.
• The developed model aids in designing and optimizing synthetic routes through predictive reaction planning.