Modelling and estimation of chemical reaction yields from high-throughput experiments.

Area of Science:

• Chemistry
• Data Science
• Chemical Engineering

Background:

• Machine learning (ML) and artificial intelligence (AI) are increasingly used to study chemical reactions.
• High-throughput experimentation (HTE) generates large datasets to understand reaction conditions, yields, and selectivities.
• Existing ML/AI methods often overlook the inherent structure within HTE datasets, potentially leading to inaccurate conclusions.

Purpose of the Study:

• To introduce a statistical modeling approach for HTE-generated chemical data.
• To develop a parameter estimation algorithm tailored for this data structure.
• To leverage the statistical model for new insights into chemical reaction mechanisms, using Buchwald-Hartwig amination as a case study.

Main Methods:

• Developed a statistical model to capture the structure of HTE data.
• Implemented a parameter estimation algorithm for the statistical model.
• Applied the model to a complex Buchwald-Hartwig amination dataset.

Main Results:

• Demonstrated that incorporating data structure knowledge yields more reliable and interpretable results than standard ML/AI.
• Obtained new mechanistic insights into the Buchwald-Hartwig amination reaction.
• Validated the applicability of the approach to diverse HTE datasets.

Conclusions:

• Leveraging the data-generating process through statistical modeling enhances the reliability and interpretability of ML/AI in chemical research.
• This approach provides a robust framework for analyzing complex chemical datasets.
• The methodology is broadly applicable to HTE data across various scientific domains beyond chemistry.