A Multi-Objective Active Learning Platform and Web App for Reaction Optimization.

Area of Science:

• Chemistry
• Chemical Engineering
• Computational Chemistry

Background:

• Multi-objective reaction optimization is crucial for chemical synthesis.
• Traditional optimization methods can be time-consuming and labor-intensive.
• High-throughput experimentation (HTE) generates large datasets but requires efficient analysis.

Purpose of the Study:

• To develop an open-source Bayesian optimization platform for multi-objective reaction optimization.
• To create a user-friendly interface for chemists to integrate optimization routines into laboratory practices.
• To demonstrate the platform's effectiveness in optimizing complex chemical reactions.

Main Methods:

• Development of an open-source Bayesian optimization platform.
• Utilizing high-throughput experimentation (HTE) and virtual screening data.
• Fine-tuning algorithm components including reaction encodings and surrogate models.
• Application to a Ni/photoredox-catalyzed enantioselective cross-electrophile coupling reaction.

Main Results:

• The Bayesian optimizer identified superior reaction conditions in significantly fewer experiments (15-24) compared to human-driven campaigns.
• The platform successfully optimized simultaneous reaction yield and enantioselectivity.
• Achieved results surpassed previous human-driven optimization efforts.

Conclusions:

• The developed open-source platform provides an efficient and accessible solution for multi-objective reaction optimization.
• The graphical user interface (GUI) lowers the barrier for chemists to utilize advanced optimization techniques.
• This approach accelerates chemical discovery and integrates computational tools into routine laboratory work.