(572j) Efficiently Exploring the Adsorption Space of Molecules in MOFs Combining the Use of Molecular Simulations, Machine Learning, and IAST

The combination of billions of possible molecules and thousands of reported Metal Organic Frameworks (MOFs) forms a vast chemical space. The limited variety of experimental isotherms that are available, however, make it very challenging to comprehensively assess adsorption-based separations in MOFs. To explore a much wider adsorption space, we carefully selected a large and diverse set of molecules and MOFs to generate molecular simulation data and train machine learning models. With the resulting models, accurate and efficient predictions of adsorption properties including Henry’s constants and adsorption isotherms of arbitrary molecules in large libraries of MOFs are possible. By combining these predictions with Ideal Adsorbed Solution Theory (IAST) we make predictions for the adsorption properties of millions of distinct binary mixtures in thousands of MOFs. We will illustrate how this approach can be used to gain insights into the characteristics of MOFs suitable for a diverse range of challenging separations.