Breaking the Boundaries of Bayesian Optimization Utilizing Continuous Chemistry Digital Twins.

Area of Science:

• Chemistry
• Chemical Engineering
• Computational Chemistry

Background:

• Bayesian Optimization (BO) is effective but requires expert knowledge.
• Current BO methods are limited by predefined design spaces.
• Optimizing chemical reactions often involves complex, multi-parameter spaces.

Purpose of the Study:

• To develop a novel chemistry optimization methodology.
• To enhance Bayesian Optimization (BO) by reducing reliance on expert knowledge.
• To autonomously expand the design space during optimization.

Main Methods:

• Developed a data analysis strategy called Breaking-the-Boundaries (BtB).
• Integrated BtB with Bayesian Optimization (BtB-BO).
• Applied BtB-BO to digital twins of chemical reactions, including nucleophilic aromatic substitution and ciprofloxacin intermediate synthesis.

Main Results:

• BtB-BO successfully expanded the design space autonomously.
• The methodology achieved optimal conditions unreachable by traditional BO with limited initial design spaces.
• BtB-BO maintained BO's optimization rate, finding solutions in under 30 iterations for complex problems.

Conclusions:

• The BtB-BO methodology significantly improves chemical process optimization.
• Autonomous design space expansion removes the need for extensive prior expert knowledge.
• BtB-BO offers a powerful tool for discovering optimal conditions in complex chemical systems.