EAGER: ADAPT: AI-based Categorization to Decipher Reaction Mechanisms from Cyclic Voltammetry

WIth support from the NSF Directorate of Mathematical and Physical Sciences, Artificial Intelligence Program, and the Division of Chemistry, Chong Liu (PI) and Quanquan Gu (co-PI) at the University of California, Los Angeles (UCLA) will apply artificial intelligence (AI) to extract mechanistic information from cyclic voltammetry experiments. Electrochemical measurement is a fundamental technique for the analysis of reaction mechanisms from synthetic organic chemistry and enzyme-catalyzed transformations.

The current practice of manual analysis of measurement results demands significant training, may introduce unconscious bias, and is not compatible with high-throughput screening in search of optimal catalysts, for example. This research project addresses the aforementioned challenge by developing AI-based algorithms that decipher experimental data for a given reaction and automatically extract mechanistic information.

The project has the potential of transforming how researchers analyze the properties of catalysts or reaction promoters and may accelerate our discovery of reactionmechanism. Both the PI and co-PI strive to encourage a broader participation among students of different research and socio-economic backgrounds. Students in this program will acquire skills and knowledge in electrochemistry, programming, machine-learning and artificial intelligence.

The interdisciplinary collaboration and the co-mentoring of students between chemistry and computer science will help to develop the next-generation AI-savvy workforce.

Combining their expertise in chemistry and computer science, Drs. Liu and Dr. Gu aim to develop an analytic procedure based on AI and machine learning that alleviates the demand of manual inspection for mechanistic analysis in cyclic voltammetry.

The team is working to establish algorithms based upon a convolutional neural network for mechanism categorization from cyclic voltammograms. The UCLA team will then test the predictability of the established algorithms for competing mechanisms for which classic manual analysis has performed poorly, and develop additional algorithms based on Bayesian optimization that suggest experimental testing conditions when available information is not sufficient for a conclusive classification.

The developed algorithms have the potential to be more sensitive and allow access to unbiased, probability-based conclusions that may not be possible with manual inspection.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.