Predicting Highly Enantioselective Catalysts Using Tunable Fragment Descriptors.

Area of Science:

• Catalysis
• Machine Learning
• Computational Chemistry

Background:

• Traditional catalyst optimization relies on qualitative assumptions and screening data.
• Existing machine learning models often require costly quantum chemical calculations or lack predictive power with simpler descriptors.
• There is a need for efficient and accurate methods for catalyst virtual screening.

Purpose of the Study:

• To develop a machine learning model for robust and efficient virtual screening of catalysts.
• To address the limitations of existing methods by utilizing readily available fragment descriptors.
• To design and validate new catalysts for asymmetric catalysis.

Main Methods:

• Developed a machine learning model utilizing fine-tuned fragment descriptors for asymmetric catalysis.
• Fragment descriptors represent cyclic or polyaromatic hydrocarbons.
• Trained the model on data with moderate selectivities.

Main Results:

• The model enabled robust and efficient virtual screening.
• New catalysts were designed theoretically based on model predictions.
• Experimentally validated catalysts demonstrated higher selectivities in asymmetric tetrahydropyran synthesis.

Conclusions:

• Fragment-based machine learning models offer a powerful and cost-efficient approach for catalyst discovery.
• This method overcomes the limitations of traditional and other computational approaches.
• The developed model facilitates the identification of highly selective catalysts for challenging chemical transformations.