Reaction Discovery Involving Digital co-Expert with a Practical Application in Atom-Economic Cycloaddition.

Area of Science:

• Organic Chemistry
• Computational Chemistry
• Chemical Informatics

Background:

• Discovering new chemical transformations is crucial for advancing chemistry.
• Traditional methods for reaction discovery are time-consuming, often requiring extensive experimental screening over months or years.
• There is a need for more efficient methods to accelerate the identification of novel chemical reactions.

Purpose of the Study:

• To develop and validate a machine learning-assisted and expert-guided pipeline for accelerating the discovery of atom-economic cycloaddition reactions.
• To demonstrate the practicality of human-AI collaboration in chemical reaction discovery.
• To significantly reduce the time required for expert screening in reaction discovery.

Main Methods:

• Generated candidate reactions from publicly available quantum chemical data.
• Filtered reactions using unsupervised machine learning and clustering to reduce redundancy.
• Employed a digital co-expert for rapid prioritization, followed by human expertise for final selection and experimental validation.

Main Results:

• Accelerated the expert screening bottleneck by approximately 180-fold (from over 1200 days to 7 days).
• Identified and experimentally confirmed two novel cycloaddition reactions within approximately one week.
• Discovered previously undescribed chemical products from the novel reactions.

Conclusions:

• The human-AI collaboration pipeline is practical, cost-effective, and compatible with existing laboratory infrastructure.
• This approach significantly accelerates the discovery of new chemical reactions by overcoming the expert screening bottleneck.
• The demonstrated workflow efficiently expands accessible chemical space by combining computational screening, machine learning, and expert knowledge.