Efficient and Accurate Charge Assignments via a Multilayer Connectivity-Based Atom Contribution (m-CBAC) Approach

Abstract

Metal−organic frameworks (MOFs) have drawn considerable attention for their potential in a variety of energy applications such as gas separations and storage. With thousands of MOFs reported and more being discovered, molecular simulations can play a critical role in facilitating the material discovery. In those calculations, accurate charge assignments to the framework atoms are essential. In this study, we expand on the connectivity-based atom contribution (CBAC) method to develop an efficient, robust, and accurate approach for charge assignments. Distinct from the original CBAC method, which uses 1st layer connectivity of a target atom, our approach, denoted as multilayer CBAC (m-CBAC), incorporates multilayer connectivity up to 2nd layers. An extensive set of ∼2700 MOFs with the density-derived electrostatic and chemical (DDEC) charges is used to train the databases. The approach assigns charges in a systematic manner, where the highest-level connectivity database (i.e., 2nd-layer connectivity) is first searched, followed by lower-level connectivity patterns until the connectivity pattern is recognized. This approach makes the charge predictions feasible to almost all MOFs. Our results show that the charges assigned using m-CBAC resemble the DDEC charges very well (Pearson coefficient of 0.988). At the same time, the m-CBAC approach is computationally efficient, which is orders of magnitude faster than quantum mechanical approaches. Also, this study demonstrates that the accurate charge assignments from m-CBAC lead to reliable predictions on the Henry coefficient of CO 2 in MOFs. Overall, the m-CBAC approach can enable fast charge assignments for MOFs with good accuracy, and a software for m-CBAC charge assignments together with charges assigned for ∼12 000 MOFs in a recently released MOF database is made available along with this work.

Monte Carlo simulations for water adsorption in porous materials: Best practices and new insights

“…66 Two charge assignment schemes were used to assign partial atomic charges to MOF framework atoms. The density-derived electrostatic and chemical (DDEC) charge assignment scheme developed by Manz et al, [67][68][69] and employed by Nazarian et al 70 for MOF structures as well as the so-called multilayer connectivity-based atom contribution (mCBAC) 71 charges developed by some of us were used. This mCBAC charge assignment scheme was used to compare between the different water models in Section 3.2, while the charges reported by Nazarian et al 70 using the DDEC scheme [67][68][69] were used in Section 3.3.…”

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

“…Approach (2) includes charge equilibration methods (QEq), 34,35 statistical machine learning models, [36][37][38][39] and nearest-neighbor-like approaches based on the chemical element and bonding environment of the atom. [40][41][42][43] Generally, approach (1) produces a more accurate electrostatic potential in the pores of the MOF but incurs a computational cost orders of magnitude greater than the cost of approach (2). Thankfully, Nazarian et al 44 performed periodic DFT calculations to obtain the electron densities in ca.…”

Parametrization of Non-Bonded Force Field Terms for Metal-Organic Frameworks Using Machine Learning Approach

“…The presented partial charge predictor slightly outperforms PACMOF [87] and MPNNs [86] in MAE: 0.0113, 0.0192, and 0.025 e -, respectively. Less representative Pearson and Spearman coefficients are given for the m-CBAC [85] approach. Their values (0.997 and 0.984, respectively) are lower than those presented in this study (0.9985 and 0.9960).…”

“…The insignificant difference may be due to distinction in featurization schemas, and more importantly, the removal of duplicate data in this study. The aforementioned approaches [84,85,86,87] have been validated by comparing values of adsorption properties calculated using ML derived and DDEC charges. Thus, Spearman rank coefficient between the CO 2 Henry coefficients computed using DDEC and ML derived charges (obtained by the m-CBAC [85] approach and MPNNs [86]) equals 0.939 and 0.96, respectively.…”

“…Several recent studies [84,85,86,87] have partially solved the fast generation of FF components using a computational approach beyond atomistic simulations. Thus, machine learning (ML) algorithms make it possible to predict partial charges in MOFs within the accuracy of the underlying DDEC approach.…”