Developing Pharmaceutically Relevant Pd-Catalyzed C-N Coupling Reactivity Models Leveraging High-Throughput...

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Organic Synthesis

Background:

• Palladium-catalyzed C-N couplings are crucial in synthesizing pharmaceuticals.
• Developing predictive models for these reactions is challenging due to complex reaction spaces.

Purpose of the Study:

• To develop and validate machine learning models for predicting the success of Pd-catalyzed C-N couplings.
• To leverage a large, de novo generated dataset for robust model training and evaluation.

Main Methods:

• Generated a large dataset (4204 unique products) using high-throughput experimentation.
• Developed novel, automation-friendly C-N coupling conditions compatible with nanomole scale reactions using LiOTMS as a base.
• Employed five distinct data-splitting strategies to rigorously assess model performance, including interpolation and extrapolation capabilities.

Main Results:

• Machine learning models demonstrated high predictive performance across all data splits, as indicated by standard evaluation metrics.
• Models accurately predicted outcomes for validation libraries not included in the training set, demonstrating strong generalization.
• The developed reaction conditions were compatible with automation and nanomole scale synthesis.

Conclusions:

• The developed machine learning models offer high accuracy in predicting Pd-catalyzed C-N coupling outcomes.
• These models can significantly improve the efficiency of medicinal chemistry campaigns by enriching successful C-N couplings.
• The study highlights the power of large-scale, de novo data generation and machine learning in accelerating drug discovery.