Fully Automatized Optimization of Ring-Opening Reactions in Lactone Derivatives via Two-Step Machine Learning.

Area of Science:

• Computational chemistry
• Materials science
• Organic chemistry

Background:

• Cyclization and cycloreversion are key processes in designing functional molecules for nanoelectronics.
• Lactones are important building blocks in molecular and nanoelectronic applications.

Purpose of the Study:

• To develop a fully automatic computational platform for designing five- and six-membered ring lactones.
• To optimize the ring-opening reaction rate of lactone derivatives.

Main Methods:

• Utilized density functional theory (DFT) and transition state theory to study reaction mechanisms.
• Developed an evolutionary algorithm for data set generation and machine learning models for property prediction.
• Identified a descriptor to correlate with reaction rates, enabling prediction from ground-state calculations.

Main Results:

• Achieved a 90% reduction in computational time using machine learning models trained on ~800 molecules.
• Generated data sets three orders of magnitude larger than the initial set, enabling broad chemical space exploration.
• Successfully correlated a simple descriptor with lactone ring-opening reaction rates.

Conclusions:

• The developed platform automates the design of lactones with optimized reaction kinetics.
• The approach significantly reduces computational cost and effort for exploring chemical space.
• The modular platform can be extended to design more complex molecular systems.