Accelerating materials discovery for water crisis: Multi-objective machine learning for atmospheric water harvesting by MOFs

Abstract Addressing the global water crisis requires efficient water supply solutions. Metal-organic frameworks (MOFs) offer promise for atmospheric water harvesting (AWH). However, many MOFs suffer from poor water stability or limited adsorption capacity. To accelerate discovery, we conceptualize structure–property relationships and develop an artificial intelligence-based multi-objective workflow that evaluates MOF water uptake at low and high relative humidity, water selectivity, and stability. A wide range of classification and regression models, hyperparameter spaces, and feature selection methods are tested, with the light gradient boosting machine (LGBM) model achieving the best performance. Results reveal that water uptake and selectivity depend mainly on structural features while chemical features dominate stability. The workflow is validated on benchmark water-harvesting MOFs and newly reported stable structures. We identify the top 100 MOFs as leading AWH candidates and propose design rules to guide experimental efforts and new research directions. The workflow is available as AquaMOF, a user-friendly software package with a web interface (https://aquamof.website/), enabling on-the-fly predictions of the AWH potential of new MOFs.