Machine learning-assisted screening of metal-organic frameworks (MOFs) for the removal of heavy metals in aqueous solution

Developing heavy metal adsorbents with high efficiency is imperative for advanced wastewater treatment. So far, the design of adsorbents has primarily relied on the experimental and molecular simulation methods, which is inefficient and time-consuming due to the vast number of potential materials. This study introduces a machine learning-assisted high-throughput screening strategy to identify optimal metal-organic frameworks (MOFs) for Pb2+ removal in aqueous solution, aiming to guide the design of high-performance MOFs. First, we extracted the structural and chemical properties of MOFs from a database containing 146,205 MOFs and developed a machine learning-guided evaluation method for MOFs. This process led to the selection of 50 high performance MOFs. Considering the effects of water, we further refined our selection to 26 water-stable MOFs by literature data and computational results. Subsequently, top-10 high-performance MOFs were identified, which exhibited high Pb2+ adsorption capacity in aqueous phase. Experimental results using screened MOFs indicated the sequence of Pb2+ adsorption as follows: HKUST-1 (top1) > ZIF-8 (ranked 156) > MOF-808 (ranked 379) > MIL-101(Fe) (ranked 582) > UiO-66 (ranked 862), further validating the effectiveness of our screening strategy. Finally, based on the shared features of the top 10 MOFs, we found that regulation of topology and the coordination of free-standing carboxyl groups in MOFs can strengthen the adsorption for Pb2+. These data-driven findings can offer more rational guidance than experimental approach for the design of novel adsorbents.