Multiobjective Optimization of Metal–Organic Framework Structural Properties and Synthesis Costs through Machine Learning

Metal-organic frameworks (MOFs) are a novel class of porous materials characterized by high surface area, high porosity, and tunable structure, possessing immense application potential. Many synthesis methods have been developed for MOFs, such as chemical, hydrothermal, mechanochemical, and microwave-assisted methods. However, their commercialization faces significant hurdles due to the high synthesis costs that mainly include the material costs of organic ligands, metal precursors, solvents, and energy consumption during prolonged reaction times. The synthesis conditions and material selections also impact the MOF structural properties, such as specific surface area and porosity. Hence, in this paper, we propose a machine learning-based multiobjective framework to identify optimal MOF synthesis conditions while achieving desired structural properties. Leveraging synthesis data from the SynMOF database and large-scale commercial raw material cost data, we developed machine learning predictive models. Pareto-optimal synthesis routes maximizing SSA while minimizing cost were identified using the nondominated sorting genetic algorithm II (NSGA-II). Subsequently, k-Nearest Neighbors (k-NN) was employed to retrieve practical optimal synthesis conditions by matching the optimized abstract features with those of metal types, solvent types, and ligand descriptors. Finally, a Bagging-integrated positive-unlabeled learning model was employed to screen solutions with a production feasibility probability exceeding 0.7. This comprehensive framework provides a strategic pathway for synthesizing high-performance, low-cost MOFs.