Accelerating single-atom ORR catalyst discovery through theory and machine learning: a critical review

The oxygen reduction reaction (ORR) is an important efficiency-determining process in fuel cells and metal-air batteries, requiring highly efficient and low-cost electrocatalysts. Single-atom catalysts (SACs) have recently emerged as a revolutionary type of catalyst for ORR reaction due to their optimized atom use, precise coordination environment, and controlled electronic structure. Theoretical simulations, especially density functional theory (DFT), have been highly influential in understanding mechanisms of ORR reaction and defining structure-activity relationships of SACs during the last few years. Nevertheless, a very substantial number of possible SAC compositions and structures represents an intrinsic difficulty of solely theoretical catalyst research. In this context, synergy of machine learning (ML) with DFT results has emerged a new paradigm with high potential for speeding up the rational design of SACs for ORR. This review presents a compilation of the latest results on the theory-guided and ML-assisted investigations on ORR active SACs. First, basic concepts of ORR, theory-based descriptors for ORR activity, selectivity, and stability at isolated metallic sites are reviewed. Next, the latest approaches based on ML, including supervised machine learning, descriptor-based approaches, high-throughput techniques, and generative modeling, which use DFT-based data to quickly estimate the ORR energetics of promising SAC structures. Particular attention is given to M–N–C catalysts, coordination asymmetry, axial ligation, frameworks described by dual descriptors, and the role of graph neural networks for the local chemical environments. Finally, we also payed special attention to the challenges, such as the lack of sufficient data, interpretability, and the aforementioned transferability and dynamic effects. Concluding, we provided future perspectives on closed-loop autonomous catalyst discovery, the theory, machine learning, and experimental validation for the development of the next generation ORR electrocatalysts. • Coordination chemistry, DFT and machine learning–driven discovery strategies for single-atom ORR catalysts have been discussed. • Descriptor engineering, linking electronic, structural, and interfacial parameters with ORR performance have been highlighted. • Recent DFT–ML case studies enabling high-throughput screening approaches for SACs are described. • The key challenges and possibilities with DFT–ML approaches for SACs are also explored.