Highly parallel optimisation of chemical reactions through automation and machine intelligence

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Machine Learning

Background:

• Optimizing chemical reactions is crucial for efficient synthesis, but complex factors like high dimensionality and experimental noise pose significant challenges.
• Automated high-throughput experimentation (HTE) accelerates data acquisition but requires sophisticated methods for effective optimization.
• Traditional experimental approaches can be time-consuming and may not fully explore complex reaction landscapes.

Purpose of the Study:

• To develop and validate a scalable machine learning (ML) framework, named Minerva, for multi-objective reaction optimization.
• To demonstrate the framework's ability to handle real-world laboratory constraints and complex chemical reactivity.
• To apply Minerva in both academic research (nickel-catalyzed Suzuki reactions) and industrial pharmaceutical process development.

Main Methods:

• Development of a scalable machine learning (ML) framework (Minerva) integrating multi-objective optimization with automated high-throughput experimentation (HTE).
• Benchmarking Minerva's performance against experimental data, focusing on efficiency in large parallel batches, high-dimensional spaces, and noisy reaction data.
• Experimental validation through a 96-well HTE campaign for nickel-catalyzed Suzuki reactions and application in optimizing active pharmaceutical ingredient (API) syntheses.

Main Results:

• Minerva demonstrated robust performance, efficiently managing large parallel batches, high-dimensional search spaces, and experimental noise.
• The framework successfully navigated complex reaction landscapes, identifying optimal conditions for a nickel-catalyzed Suzuki reaction with unexpected reactivity.
• In pharmaceutical process development, Minerva optimized two API syntheses, achieving >95 area percent yield and selectivity for both Ni-catalyzed Suzuki and Pd-catalyzed Buchwald-Hartwig reactions.

Conclusions:

• Minerva provides a scalable and efficient ML-driven solution for multi-objective reaction optimization using HTE.
• The framework effectively addresses challenges in non-precious metal catalysis and complex reaction optimization.
• Minerva's successful application in pharmaceutical process development highlights its potential for improving scaled-up chemical manufacturing.