Muhammad Harussani Moklis, Cries Avian, Cheng Shuo, Sasipa Boonyubol, Jenq-Shiou Leu, Koichi Mikami, Jeffrey S. Cross
Glycerol, a major by-product generated in abundance during biodiesel production, presents a valuable opportunity for conversion into high-value chemicals through electrochemical processes. One of the main challenges in advancing glycerol electroreduction is the inefficiency of current electrocatalyst discovery methods to recognize materials tailored for specific cathodic reaction. Conventional trial-and-error approaches struggle to efficiently navigate the vast chemical design space for optimizing electrocatalyst properties. Here, we employ a generative adversarial network (GAN) to discover new hypothetical low-cost non-noble metallic electrocatalysts favoring cathodic reactions in glycerol electrocatalytic reduction (ECR). Trained on a curated dataset of over 5000 thermodynamically stable mono-, bi-, and trimetallic compounds from the Materials Project (MP) database, our GAN architecture generates 400,000 hypothetical candidates not existing in the training dataset with a uniqueness of 99.94 %, while adhering to chemical validity, thermodynamic feasibility, and electrochemical property constraints. Notably, the GAN learns implicit chemical rules despite no explicit enforcement, producing chemically valid material compositions. Further conditional screening identifies 18 top candidates, with metallic compounds significantly outnumbering metallic oxides–highlighting the natural favorability of metallic systems for electroreduction. Among the top candidates, Co-Zr-X (X = Ba, Ti) trimetallics emerge, exhibiting promise for suppressing hydrogen evolution reaction (HER) and enhancing selective glycerol ECR. These findings underscore the potential of GAN-based generative design in accelerating electrocatalyst discovery, offering a data-driven pathway to sustainable biodiesel by-product utilization. © 2025 The Author(s)
Energy Science and Engineering, Department of Transdisciplinary Science and Engineering, Institute of Science Tokyo, 2-12-1, Ookayama, Tokyo, Meguro-ku, 152-8550, Japan; Department of Electronic and Computer Engineering, National Taiwan University of Science and Technology, Taipei 106, Taiwan; Department of Electrical Engineering, Universitas Brawijaya, Malang, Jawa Timur, 65145, Indonesia
