Metal-organic frameworks (MOFs) are a class of crystalline materials composed of metal nodes or clusters connected via semi-rigid organic linkers. Owing to their high surface area, porosity, and tunability, MOFs have received significant attention for numerous applications such as gas separation and storage. Atomistic simulations and data-driven methods (e.g., machine learning) have been successfully employed to screen large databases and successfully develop new experimentally synthesized and validated MOFs for CO2 capture. To enable data-driven materials discovery for any application, the first (and arguably most crucial) step is database curation. This work introduces the ab initio REPEAT charge MOF (ARC-MOF) database. This is a database of ~280,000 MOFs which have been either experimentally characterized or computationally generated, spanning all publicly available MOF databases. A key feature of ARC-MOF is that it contains DFT-derived electrostatic potential fitted partial atomic charges for each MOF. Additionally, ARC-MOF contains pre-computed descriptors for out-of-the-box machine learning applications. An in-depth analysis of the diversity of ARC-MOF with respect to the currently mapped design space of MOFs was performed – a critical, yet commonly overlooked aspect of previously reported MOF databases. Using this analysis, balanced subsets from ARC-MOF for various machine learning purposes have been identified. Other chemical and geometric diversity analyses are presented, with an analysis on the effect of charge assignment method on atomistic simulation of gas uptake in MOFs.

This is a database of ~280,000 MOFs which have been either experimentally characterized or computationally generated, spanning all publicly available MOF databases. DFT-derived REPEAT charges, adsorption data, and various descriptors are available for all MOFs. all_structures_1.tar.gz and all_structures_2.tar.gz – these are the cif files that were considered to compose the “entire known design space” of MOFs, with any bad structures removed (split into two separate tarballs since it is a lot of data). ARCMOF_20220610.tar.gz – these are all of the cif files with REPEAT charges composing ARC-MOF. flig-clusters.csv, func-clusters.csv, geo-clusters.csv, mc-clusters.csv – Each file indicates for each MOF which cluster it belongs to, and whether the MOF is present in ARC-MOF. This is done for each "type" of MOF chemistry and for the geometric properties. Clusters with a negative value indicate the MOF does not belong to any cluster (i.e., it is assumed to be "unique"). all_topology_lists.csv – a csv file containing the topology reported by the filename of applicable structures, and the topology reported by CrystalNets.jl ML_test_set.tar.gz – these are the cif files (with REPEAT charges) of the MOFs in the diverse-mc subset, but missing from ARC-MOF (for the purposes of a ML test set for the prediction of metal charges). geometric_properties.csv – a csv file containing geometric descriptors computed for this study for all MOFs. The csv file also indicates which MOFs are present in ARC-MOF, and the order in which they were chosen for the farthest point sampling (up to 100K MOFs). RACs.csv – See geometric_properties.csv description. Same type of file, but with the RAC descriptors. RDFs.csv – The RDFs for each MOF, using several atomic properties. Some atomic properties are not available for all elements. In the cases where the atomic property is not available for a particular structure, no value is assigned. methane.csv, methane_purification-CH4.csv, methane_purification_CO2.csv, post_comb_vsa-CO2.csv, post_comb_vsa-N2.csv, pre_comb_4040-CO2.csv, pre_comb_4040-H2.csv, landfill-CH4.csv, landfill-CO2.csv – these are csv files of the raw uptake data and various temperature, pressure conditions (with standard deviations) for each gas separation process specified in the file overall_process.csv. overall_process.csv – This is a csv file of the adsorption properties of the MOFs. Particularly, the csv files contain the working capacity (mmol/g_working_capacity) and selectivity of each MOF for each of the five process conditions. mc-diverse-set.csv, func-diverse-set.csv – csv files containing which MOFs are present in each diverse set (from farthest point sampling of the MOFs based on either their functional group chemistry or metal chemistry). The file indicates which MOFs are present in ARC-MOF and which are not. Version history of repository: v2 -- added file: "all_topology_lists.csv" v3 -- added file: "ML_test_set.tar.gz" v4 -- replaced file: "ML_test_set.tar.gz". Originally incorrect repository of cifs

Metal-organic frameworks (MOFs) have garnered interest as potential solid sorbent materials for postcombustion CO2 capture. With a seemingly infinite design space, high-throughput computational screening of MOFs has developed into an effective tool for the development of new materials. In this work, machine learning (ML) has been used to develop accurate quantitative structure-property relationship (QSPR) models to rapidly predict the CO2 working capacity and CO2/N-2 selectivity at the low-pressure conditions relevant to postcombustion carbon capture (0.15 bar CO2, 0.85 bar N-2). A database of over 340 000 MOFs constructed from hundreds of types of building units arranged in over 1000 net topologies was used to train and test the models. Neural network ML models were optimized using six geometric descriptors along with three so-called chemical descriptors, namely, the atomic property-weighted radial distribution function (AP-RDF) and some variants thereof, the bag-of-atoms, and the chemical motif density descriptors. The ML models built using geometric descriptors alone resulted in test set correlation R-2 values of only 0.71 and 0.75 for CO2 working capacity and CO2/N-2 selectivity, respectively. ML models built with a single type of chemical descriptor all outperformed the geometry-only models giving R-2 values ranging from 0.83 to 0.94 with the AP-RDF model being the most accurate. Overall, the best model was built using a combination of AP-RDF, chemical motif, and geometric descriptors (R-2 = 0.96 when predicting the CO2 working capacity and R-2 = 0.95 for the selectivity). To date, these are the most accurate ML models for predicting low-pressure gas uptake of MOFs. The combined model was able to capture 994 of the true top 1000 MOFs (from a test set of similar to 70 000) within the top 5000 MOFs as predicted by the model with CO, working capacity as the target. Thus, if the ML model were used to prescreen materials for more compute intensive GCMC simulations, then it would result in a greater than 10 times speed up while still capturing >99% of high-performing materials. These results highlight the importance of chemical descriptors in predicting low-pressure gas adsorption properties in nanoporous materials.