Incorporating Synthetic Accessibility in Drug Design: Predicting Reaction Yields of Suzuki Cross-Couplings by...

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Machine Learning

Background:

• Predicting synthetic yields is a major challenge in drug design.
• Improved yield prediction can reduce material waste and accelerate the design-make-test-analyze (DMTA) cycle.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting Suzuki coupling reaction yields.
• To assess the performance of DFT-derived features and Morgan fingerprints for yield prediction.

Main Methods:

• Utilized a medicinal chemistry library dataset from AbbVie.
• Developed machine learning models incorporating density functional theory (DFT)-derived features and Morgan fingerprints.
• Compared model performance against one-hot encoded baselines and expert medicinal chemists.

Main Results:

• The combination of DFT-derived features and Morgan fingerprints showed superior performance.
• The model demonstrated modest generalization to unseen reactant structures.
• The machine learning model predicted reaction success and yields with higher accuracy than expert chemists.

Conclusions:

• Machine learning models can effectively predict Suzuki coupling reaction yields.
• This approach enhances synthesis efficiency by suggesting alternative building blocks with higher predicted yields.
• The developed models can significantly aid medicinal chemists in optimizing drug design workflows.