Robust Machine Learning Models for Predicting High CO2 Working Capacity and CO2/H2 Selectivity of Gas Adsorption in Metal Organic Frameworks for Precombustion Carbon Capture

This work is devoted to the development of quantitative structure–property relationship (QSPR) models using machine learning to predict CO2 working capacity and CO2/H2 selectivity for precombustion carbon capture using a topologically diverse database of hypothetical metal–organic framework (MOF) structures (358 400 MOFs, 1166 network topologies). Such a diversity of the networks topology is much higher than previously used (<20 network topologies) for rapid and accurate recognition of high-performing MOFs for other gas-separation applications. The gradient boosted trees regression method allowed us to use 80% of the database as a training set, while the rest was used for the validation and test set. The QSPR models are first built using purely geometric descriptors of MOFs such as gravimetric surface area and void fraction. Additional models which account for chemical features of MOFs are constructed using atomic property weighted radial distribution functions (AP-RDFs) with a novel normalization to accommodate the size diversity of the MOF database. It is shown that the best models for CO2 working capacities (R2 = 0.944) and CO2/H2 selectivities (R2 = 0.872) are built from a combination of six geometric descriptors and three AP-RDF descriptors. However, more important is that our QSPR models can identify top 1000 high-performing MOFs in just top 3000 or 5000 MOFs. This work shows that QSPR modeling can account for the topological diversity of MOFs and accelerate the screening for top-performing MOFs for precombustion carbon capture.