A machine-learning tool to predict substrate-adaptive conditions for Pd-catalyzed C-N couplings

Area of Science:

• Organic Chemistry
• Computational Chemistry
• Chemical Engineering

Background:

• Machine learning (ML) offers significant potential for accelerating the optimization of reaction conditions in chemical synthesis.
• Palladium-catalyzed carbon-nitrogen (C-N) couplings are crucial transformations in organic synthesis, but identifying optimal conditions can be challenging.
• Developing predictive models requires comprehensive experimental datasets covering diverse reactant pairings and reaction parameters.

Purpose of the Study:

• To develop a machine-learning tool that provides substrate-adaptive reaction conditions for palladium-catalyzed C-N couplings.
• To create a systematic workflow for actively learning and improving predictive models for chemical transformations.
• To demonstrate the tool's efficacy in predicting high-yield reaction conditions for novel reactants.

Main Methods:

• Generation of a diverse experimental dataset exploring various reactant pairings and reaction conditions for C-N couplings.
• Training neural network models using a systematic experimental design process to actively learn the scope of C-N couplings.
• Experimental validation of the ML model predictions with out-of-sample reactants.

Main Results:

• The developed ML models demonstrated strong performance in predicting reaction conditions.
• Ten target products were successfully isolated in yields exceeding 85% using predicted conditions for challenging, out-of-sample reactants.
• The workflow showed continuous improvement in prediction capability as more data was incorporated.

Conclusions:

• The presented tool effectively accelerates the identification of reaction conditions for Pd-catalyzed C-N couplings.
• The substrate-adaptive approach enhances the efficiency and scope of chemical synthesis.
• The continuous learning capability ensures ongoing improvement and broader applicability of the tool.