Summary:
UCLA researchers in the Department of Chemistry and Biochemistry developed a custom-designed deep learning architecture for object detection and mechanisms classification to streamline the experimental design of cyclic voltammograms.
Background:
Cyclic voltammetry, a widely used electrochemical test, serves as a crucial tool in understanding chemical behaviors in various environments. By manipulating electron addition and removal, researchers instigate an electrochemical reaction, providing insights into how a chemical may respond. This information can be subsequently used to incorporate chemicals into appropriate applications and have a comprehensive understanding of what the expected behavior will be. An important phase of cyclic voltammetry is mechanism assignment, which delineates how molecules interact under electrical stimulation. Traditionally, researchers perform mechanism classification through subjective visual analysis, identifying redox events, or transfer of electrons between a molecule and electrode surface, in the voltammogram through object detection of graphs’ peaks and valleys. They then assign the mechanism based on established patterns and theories to specific groups.
However, this method is prone to human error, particularly in distinguishing between broad and nuanced mechanisms. This process could deeply benefit from a deep-learning model to offer an automated route to successfully detect and assign mechanisms. Existing machine learning models rely on certain prior knowledge or assumptions and require human identification of critical characteristics, limiting the potential of these models. Therefore, there is a critical gap for a deep-learning model capable of object detection and mechanism classification in multi-redox cyclic voltammograms that can operate with minimal prior knowledge.
Innovation:
UCLA researchers have developed a deep-learning model, EchemNet, capable of object detection and mechanism classification for multi-redox cyclic voltammograms with minimal a priori information. This model was based on the well-established Faster R-CNN (Regional Convolutional Neural Network) architecture, widely used for recognition and classification of two-dimensional images across diverse domains. EchemNet introduces a novel approach to recognize one-dimensional images. Trained on simulated multi-redox voltammograms, EchemNet utilizes object detection technology to identify peaks, valleys, and other critical characteristics within the voltammogram and subsequently assign each mechanism categorized into one of 8 different reaction mechanisms for up to 4 redox events. Notably, EchemNet accomplishes object detection and mechanisms classification without relying on any prior data inputs or knowledge. By automating this process, EchemNet will allow for more precise and quicker classification of various electrochemical reactions, streamlining investigations across diverse applications.
Potential Applications:
• Electrochemical Mechanism Classification
• Machine Learning Systems
• Optimization of Energy Storage Systems, such as battery performance or quality control
• Development of electrochemical sensors for healthcare diagnostics, such as biosensors, or environmental monitoring
• Material science characterization and material application assessments
Advantages:
• Increased accuracy and efficiency of electrochemical mechanism classification
• Ability to classify reactions with no a priori knowledge or assumptions
• Eliminating the need for human object detection and classification entirely
• Ability to classify reactions with more than 1 redox event, increasing the versatility of the model
Development-To-Date:
Patent application has been submitted.
Related Papers and Patents:
1. Hoar B, Zhang W, Chen Y, Sun J, Sheng H, Zhang Y, et al. Object-detecting deep learning for mechanism discernment in multi-redox cyclic voltammograms. ChemRxiv. 2023; doi:10.26434/chemrxiv-2023-r2v1k Note: This content is a preprint and has not been peer-reviewed.
2. Hoar, B. B.; Zhang, W.; Xu, S.; Deeba, R.; Costentin, C.; Gu, Q.; Liu, C. Electrochemical Mechanistic Analysis from Cyclic Voltammograms Based on Deep Learning. ACS Measurement Science Au 2022, 2 (6), 595-604. DOI: 10.1021/acsmeasuresciau.2c00045.
Reference: UCLA Case No. 2024-088