Next-generation MOFs for atmospheric water harvesting: the role of machine learning techniques

Abstract

Atmospheric Water Harvesting (AWH) using Metal-Organic Frameworks (MOFs) has emerged as a highly promising approach to mitigate water scarcity, especially in arid and semi-arid regions. The development of high-performance MOFs for AWH hinges on materials that exhibit optimal water uptake capacity, rapid adsorption-desorption kinetics, and robust hydrolytic stability. However, the structural complexity of MOFs and the inefficiencies of traditional experimental screening have made data-driven approaches, particularly machine learning (ML), increasingly indispensable for accelerating materials discovery. Among the ML techniques applied to MOF-based AWH, models such as Random Forest (RF), Random Forest Regression (RFR), Neighbor Component Analysis (NCA), Genetic Algorithms (GA), and Machine-Learned Atomistic Cluster Expansion (MACE) have demonstrated outstanding predictive performance. These models are especially valued for their ability to capture non-linear dependencies, improve interpretability, and optimize design strategies across diverse application domains. This review presents a comprehensive analysis of ML-assisted MOF discovery for AWH, focusing on the roles of explicit and latent descriptors, evaluation metrics, dataset curation challenges, and comparative model performance. By emphasizing the superior predictive capabilities of RF, MACE, NCA, RFR, and GA, this work highlights the transformative potential of ML in driving the rational design of next-generation MOFs for efficient and scalable AWH.