Optimizing Vanadium-Catalyzed Epoxidation Reactions: Machine-Learning-Driven Yield Predictions and Data Augmentation.

Area of Science:

• Catalysis
• Chemical Engineering
• Data Science

Background:

• Catalytic epoxidations are crucial for synthesizing valuable compounds.
• Optimizing these reactions is essential for industrial applications.

Purpose of the Study:

• To develop a supervised machine learning (ML) model for predicting vanadium-catalyzed epoxidation reaction yields.
• To identify key chemical descriptors for optimizing epoxidation reactions and catalyst design.

Main Methods:

• A supervised ML model was developed and trained on 273 experimental epoxidation reactions.
• Data augmentation techniques were used to handle experimental variability.
• Descriptor analysis was performed to understand model predictions.

Main Results:

• The ML model achieved a predictive R² test score of 90%.
• The model demonstrated a mean absolute yield prediction error of 4.7%.
• Key experimental and chemical descriptors influencing catalytic predictions were identified.

Conclusions:

• The developed ML model accurately predicts epoxidation yields and offers high explainability.
• This work highlights the potential of data science in advancing catalytic epoxidation research and catalyst optimization.