The central aim of this project is to develop modeling tools that vastly increase the diversity of molecules for which reliable adsorption isotherms in nanoporous materials are available.The scope of the work is shown in Fig. 1.Within the project, molecular simulation data for a diverse set of examples is being combined with machine-learning (ML) methods to rapidly predict adsorption equilibria in crystalline nanoporous materials (primarily MOFs).Developing models that can rapidly be applied to very large and diverse collections of molecules (e.g.thousands of different species) represents a key means in which modeling could dramatically accelerate the scope of MOFs or similar materials for chemical separations in ways that cannot be accomplished experimentally.Rather than predicting the entire adsorption isotherm for individual molecules or adsorbed mixtures, we focused instead on predicting the Henry's constants for each species in each MOF.These quantities are directly related to adsorption selectivity because in the dilute limit adsorption selectivity is, without approximation, the ratio of Henry's constants of the adsorbing species.Just as important, previous empirical and ML-based models of adsorption in MOFs by us and others have shown that the Henry's regime is considerably more challenging to predict than the saturation regime.By comparing ML-based predictions and direct molecular simulations for multiple mixtures of near-azeotropic molecules (that is, pairs of molecules with very similar boiling points), it was shown that accurate prediction of single-component adsorption in the dilute loading limit is essential to making accurate mixture adsorption predictions 1 .