Predicting reaction performance in C-N cross-coupling using machine learning.

Area of Science:

• Chemistry
• Computational Chemistry
• Chemical Engineering

Background:

• Machine learning (ML) is increasingly vital in scientific research.
• Predicting chemical reaction outcomes is crucial for optimizing synthetic processes.
• High-throughput experimentation (HTE) generates large datasets for ML model training.

Purpose of the Study:

• To investigate the utility of ML for predicting synthetic reaction performance.
• To compare ML algorithms against traditional methods in chemical space.
• To assess the applicability of ML models for facilitating synthetic methodology adoption.

Main Methods:

• Extracted atomic, molecular, and vibrational descriptors for reaction components.
• Utilized a palladium-catalyzed Buchwald-Hartwig cross-coupling reaction as a model system.
• Employed random forest and linear regression algorithms for predictive modeling using HTE data.

Main Results:

• The random forest algorithm significantly outperformed linear regression in predicting reaction yield.
• The ML model demonstrated robust performance even with sparse training datasets.
• Successful out-of-sample predictions were achieved, validating the model's generalizability.

Conclusions:

• Machine learning, specifically random forest, can accurately predict synthetic reaction performance.
• This approach is valuable for navigating multidimensional chemical space and optimizing reactions.
• ML-driven predictions can accelerate the adoption and development of new synthetic methods.