Machine learning and molecular descriptors enable rational solvent selection in asymmetric catalysis.

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Process Development

Background:

• Solvent selection is a critical bottleneck in optimizing chemical reactions.
• Developing predictive models for solvent effects requires significant experimental data and expertise.

Purpose of the Study:

• To develop an automated, hybrid mechanistic-machine learning approach for rational solvent selection.
• To optimize a Rh(CO)2(acac)/Josiphos catalyzed asymmetric hydrogenation for high conversion and diastereomeric excess.

Main Methods:

• Calculated molecular and reaction-specific descriptors for 459 solvents.
• Trained Gaussian process surrogate models on experimental data for conversion and diastereomeric excess.
• Employed multi-objective optimization and Bayesian optimization for solvent and condition screening.

Main Results:

• Identified optimal solvents and solvent mixtures for improved reaction outcomes.
• Achieved a cross-validation correlation coefficient of 0.84 for the conversion prediction model.
• Demonstrated a genetic programming approach for selecting appropriate machine learning models.

Conclusions:

• The hybrid approach enables efficient and automated solvent selection in process development.
• This methodology supports the integration of machine learning into future robotic chemical laboratories.
• Predictive modeling significantly accelerates the optimization of complex catalytic reactions.