Message Passing Neural Networks for Partial Charge Assignment to Metal–Organic Frameworks

Abstract

Virtual screenings can accelerate and reduce the cost of discovering metal-organic frameworks (MOFs) for their applications in gas storage, separation, and sensing. In molecular simulations of gas adsorption/diffusion in MOFs, the adsorbate-MOF electrostatic interaction is typically modeled by placing partial point charges on the atoms of the MOF. For the virtual screening of large libraries of MOFs, it is critical to develop computationally inexpensive methods to assign atomic partial charges to MOFs that accurately reproduce the electrostatic potential in their pores. Herein, we design and train a message passing neural network (MPNN) to predict the atomic partial charges on MOFs under a charge neutral constraint. A set of ca. 2,250 MOFs labeled with high-fidelity partial charges, derived from periodic electronic structure calculations, serves as training examples. In an end-to-end manner, from charge-labeled crystal graphs representing MOFs, our MPNN machine-learns features of the local bonding environments of the atoms and learns to predict partial atomic charges from these features. Our trained MPNN assigns high-fidelity partial point charges to MOFs with orders of magnitude lower computational cost than electronic structure calculations. To enhance the accuracy of virtual screenings of large libraries of MOFs for their adsorption-based applications, we make our trained MPNN model and MPNN-charge-assigned computation-ready, experimental MOF structures publicly available. File list (2) download file view on ChemRxiv Message_passing_networks_to_learn_MOF_charges.pdf (3.62 MiB) download file view on ChemRxiv Message_passing_networks_to_learn_MOF_charges_SI.... (1.55 MiB)

Machine learning the quantum-chemical properties of metal–organic frameworks for accelerated materials discovery

“…To date, the most relevant studies focused on training ML models to predict the quantum-chemical properties of MOFs are those of Raza et al 23 and Korolev et al, 24 who independently developed ML models that can predict the partial atomic charges of MOFs in the Computation-Ready, Experimental (CoRE) MOF database. 25,26 Beyond these fundamental studies on partial charge prediction, however, there remains a significant gap in the literature, particularly for the discovery of MOFs with desired electronic structure properties.…”

Predicting chemical shifts with graph neural networks

“…The data was small organic molecules and so the model tasks was less focused on macromolecules and conformational effects than this work. Examples of others work using message passing GNNs in chemistry include Raza et al 52 who predicted partial charges of metal organic frameworks, the original message passing paper byGilmer et al 31 which predicted energies of molecules, and St. John et al 53 who predicted bond disassociation energies. There are also first-principles methods for computing NMR chemical shifts, however we do not compare with these since their computational speed and accuracy are not comparable with empirical methods [54][55][56] .…”

Performance-Based Screening of Porous Materials for Carbon Capture

“…More recently, new charge assignment schemes have been developed based on machine learning (ML) techniques where the ML model is trained on a collection of high-quality DFT-derived charges such as DDEC. 286 , 287 An example of these models is developed by Kancharlapalli et al 286 for MOFs and was shown to be transferable to other porous materials such as zeolites and porous molecular crystals. The ML-based charge assignment schemes are more beneficial for screening of large databases of porous materials where application of DFT-derived partial charges such as REPEAT or DDEC can be computationally very expensive.…”