Minimal Basis Iterative Stockholder: Atoms in Molecules for Force-Field Development

A comprehensive electron wavefunction analysis toolbox for chemists, Multiwfn

Analysis of electron wavefunction is a key component of quantum chemistry investigations and is indispensable for the practical research of many chemical problems. After more than ten years of active development, the wavefunction analysis program Multiwfn has accumulated very rich functions, and its application scope has covered numerous aspects of theoretical chemical research, including charge distribution, chemical bond, electron localization and delocalization, aromaticity, intramolecular and intermolecular interactions, electronic excitation, and response property. This article systematically introduces the features and functions of the latest version of Multiwfn and provides many representative examples. Through this article, readers will be able to fully understand the characteristics and recognize the unique value of Multiwfn. The source code and precompiled executable files of Multiwfn, as well as the manual containing a detailed introduction to theoretical backgrounds and very rich tutorials, can all be downloaded for free from the Multiwfn website (http://sobereva.com/multiwfn).

A physics-aware neural network for protein-ligand interactions with quantum chemical accuracy

Quantifying intermolecular interactions with quantum chemistry (QC) is useful for many chemical problems, including understanding the nature of protein-ligand interactions. Unfortunately, QC computations on protein-ligand systems are too computationally expensive for most use cases. The flourishing field of machine-learned (ML) potentials is a promising solution, but it is limited by an inability to easily capture long range, non-local interactions. In this work we develop an atomic-pairwise neural network (AP-Net) specialized for modeling intermolecular interactions. This model benefits from a number of physical constraints, including a two-component equivariant message passing neural network architecture that predicts interaction energies via an intermediate prediction of monomer electron densities. The AP-Net model is trained on a comprehensive dataset composed of paired ligand and protein fragments. This model accurately predicts QC-quality interaction energies of protein-ligand systems at a computational cost reduced by orders of magnitude. Applications of the AP-Net model to molecular crystal structure prediction are explored, as well as limitations in modeling highly polarizable systems.

CuGBasis: High-performance CUDA/Python library for efficient computation of quantum chemistry density-based descriptors for larger systems

CuGBasis is a free and open-source CUDA®/Python library for efficient computation of scalar, vector, and matrix quantities crucial for the post-processing of electronic structure calculations. CuGBasis integrates high-performance Graphical Processing Unit (GPU) computing with the ease and flexibility of Python programming, making it compatible with a vast ecosystem of libraries. We showcase its utility as a Python library and demonstrate its seamless interoperability with existing Python software to gain chemical insight from quantum chemistry calculations. Leveraging GPU-accelerated code, cuGBasis exhibits remarkable performance, making it highly applicable to larger systems or large databases. Our benchmarks reveal a 100-fold performance gain compared to alternative software packages, including serial/multi-threaded Central Processing Unit and GPU implementations. This paper outlines various features and computational strategies that lead to cuGBasis's enhanced performance, guiding developers of GPU-accelerated code.

DASH properties: Estimating atomic and molecular properties from a dynamic attention-based substructure hierarchy

Recently, we presented a method to assign atomic partial charges based on the DASH (dynamic attention-based substructure hierarchy) tree with high efficiency and quantum mechanical (QM)-like accuracy. In addition, the approach can be considered "rule based"-where the rules are derived from the attention values of a graph neural network-and thus, each assignment is fully explainable by visualizing the underlying molecular substructures. In this work, we demonstrate that these hierarchically sorted substructures capture the key features of the local environment of an atom and allow us to predict different atomic properties with high accuracy without building a new DASH tree for each property. The fast prediction of atomic properties in molecules with the DASH tree can, for example, be used as an efficient way to generate feature vectors for machine learning without the need for expensive QM calculations. The final DASH tree with the different atomic properties as well as the complete dataset with wave functions is made freely available.

On the Theoretical Quantification of Radii of Atoms in Molecules

Despite the fundamental importance of radii of atoms in molecules for numerous applications in physics and chemistry, comprehensive methods for their theoretical evaluation are still scarce. Here, we present quantum chemistry-based approaches for evaluation of radii of atoms in molecules and assess their robustness by studying the agreement of van der Waals and solvent-excluded surfaces constructed by them with reference molecular surfaces. By studying a large data set of 1235 molecules, it is shown that estimation of radii via effective and free atomic volumes can accurately take the dependence of atomic radii on the chemical environment into account.

Four-Dimensional Scaling of Dipole Polarizability: From Single-Particle Models to Atoms and Molecules

Scaling laws enable the determination of physicochemical properties of molecules and materials as a function of their size, density, number of electrons or other easily accessible descriptors. Such relations can be counterintuitive and nonlinear, and ultimately yield much needed insight into quantum mechanics of many-particle systems. In this work, we show on the basis of single-particle models, multielectron atoms and molecules that the dipole polarizability of quantum systems is generally proportional to the fourth power of a characteristic length, computed from the ground-state wave function. This four-dimensional (4D) scaling is independent of the ratio of bound-to-bound and bound-to-continuum electronic transitions and applies to many-electron atoms when a correlated length metric is used. Finally, this scaling law is applied to predict the polarizability of molecules by electrostatically coupled atoms-in-molecules approach, obtaining approximately 8% absolute and relative accuracy with respect to hybrid density functional theory (DFT) on the QM7-X data set of organic molecules, providing an efficient and scalable model for the anisotropic polarizability tensors of extended (bio)molecules.

FeNNol: An efficient and flexible library for building force-field-enhanced neural network potentials

Neural network interatomic potentials (NNPs) have recently proven to be powerful tools to accurately model complex molecular systems while bypassing the high numerical cost of ab initio molecular dynamics simulations. In recent years, numerous advances in model architectures as well as the development of hybrid models combining machine-learning (ML) with more traditional, physically motivated, force-field interactions have considerably increased the design space of ML potentials. In this paper, we present FeNNol, a new library for building, training, and running force-field-enhanced neural network potentials. It provides a flexible and modular system for building hybrid models, allowing us to easily combine state-of-the-art embeddings with ML-parameterized physical interaction terms without the need for explicit programming. Furthermore, FeNNol leverages the automatic differentiation and just-in-time compilation features of the Jax Python library to enable fast evaluation of NNPs, shrinking the performance gap between ML potentials and standard force-fields. This is demonstrated with the popular ANI-2x model reaching simulation speeds nearly on par with the AMOEBA polarizable force-field on commodity GPUs (graphics processing units). We hope that FeNNol will facilitate the development and application of new hybrid NNP architectures for a wide range of molecular simulation problems.

Computational Simulations of Metal-Organic Frameworks to Enhance Adsorption Applications

Metal-organic frameworks (MOFs), renowned for their exceptional porosity and crystalline structure, stand at the forefront of gas adsorption and separation applications. Shortly after their discovery through experimental synthesis, computational simulations quickly become an important method in broadening the use of MOFs by offering deep insights into their structural, functional, and performance properties. This review specifically addresses the pivotal role of molecular simulations in enlarging the molecular understanding of MOFs and enhancing their applications, particularly for gas adsorption. After reviewing the historical development and implementation of molecular simulation methods in the field of MOFs, high-throughput computational screening (HTCS) studies used to unlock the potential of MOFs in CO2 capture, CH4 storage, H2 storage, and water harvesting are visited and recent advancements in these adsorption applications are highlighted. The transformative impact of integrating artificial intelligence with HTCS on the prediction of MOFs' performance and directing the experimental efforts on promising materials is addressed. An outlook on current opportunities and challenges in the field to accelerate the adsorption applications of MOFs is finally provided.

Anion Photoelectron Imaging Spectroscopy of C6F5X- (X = F, Cl, Br, I)

The photoelectron (PE) spectra of C6F5X- (X = Cl, Br, I) and computational results on the anions and neutrals are presented and compared to previously reported results on C6F6- [McGee, C. J. J. Phys. Chem. A 2023, 127, 8556-8565.]. The spectra all exhibit broad, vibrationally unresolved detachment transitions, indicating that the equilibrium structures of the anions are significantly different from the neutrals. The PE spectrum of C6F5Cl- exhibits a parallel photoelectron angular distribution (PAD), similar to that of the previously reported C6F6- spectrum, while the PE spectra of C6F5Br- and C6F5I- have isotropic PADs, and also exhibit a prominent X- PE feature due to photodissociation of C6F5X- resulting in X- formation. Identification of the C6F5X- detachment transition origins, which is equivalent to the neutral electron affinity (EA), in all three cases is difficult, since the broadness of the detachment feature is accompanied by vanishingly small detachment cross section near the origin. Upper limits on the EAs were determined to be 1.70 eV for C6F5Cl, 2.10 eV for C6F5Br, and 2.00 eV for C6F5I, all significantly higher than the 0.76 eV upper limit determined for C6F6 with the same experiment. The broad detachment transitions are consistent with computational results, which predict very large differences between the neutral and anionic C-X (X = Cl, Br, I) bond lengths. Based on differences between the MBIS atom charges in the anions and neutrals, the excess charge in the anion is on the unique C atom and X, in contrast to the nonplanar C2v structured C6F6- anion, for which the charge is delocalized over the molecule. In C6F5Cl-, the C-Cl bond is predicted to be bent out of the plane, while both C6F5Br- and C6F5I- are predicted to be planar on average. The impact of the interruption of the symmetry in the hexafluorobenzene neutral and anion on the molecular and electronic structure of C6F5X/C6F5X- is considered, as well as the possible dissociative state leading to X- (X = Br, I) formation, and the nature of the C-X bond.

The Separability Problem in Molecular Quantum Systems: Information-Theoretic Framework for Atoms in Molecules

Even though molecules are fundamentally quantum entities, the concept of a molecule retains certain classical attributes concerning its constituents. This includes the empirical separability of a molecule into its three-dimensional, rigid structure in Euclidean space, a framework often obtained through experimental methods like X-Ray crystallography. In this work, we delve into the mathematical implications of partitioning a molecule into its constituent parts using the widely recognized Atoms-In-Molecules (AIM) schemes, aiming to establish their validity within the framework of Information Theory concepts. We have uncovered information-theoretical justifications for employing some of the most prevalent AIM schemes in the field of Chemistry, including Hirshfeld (stockholder partitioning), Bader's (topological dissection), and the quantum approach (Hilbert's space definition). In the first approach we have applied the generalized principle of minimum relative entropy derived from the Sharma-Mittal two-parameter functional, avoiding the need for an arbitrary selection of reference promolecular atoms. Within the ambit of topological-information partitioning, we have demonstrated that the Fisher information of Bader's atoms conform to a comprehensive theory based on the Principle of Extreme Physical Information avoiding the need of employing the Schwinger's principle, which has been proven to be problematic. For the quantum approach we have presented information-theoretic justifications for conducting Löwdin symmetric transformations on the density matrix to form atomic Hilbert spaces generating orthonormal atomic orbitals with maximum occupancy for a given wavefunction.

Developing a Differentiable Long-Range Force Field for Proteins with E(3) Neural Network-Predicted Asymptotic Parameters

Accurately describing long-range interactions is a significant challenge in molecular dynamics (MD) simulations of proteins. High-quality long-range potential is also an important component of the range-separated machine learning force field. This study introduces a comprehensive asymptotic parameter database encompassing atomic multipole moments, polarizabilities, and dispersion coefficients. Leveraging active learning, our database comprehensively represents protein fragments with up to 8 heavy atoms, capturing their conformational diversity with merely 78,000 data points. Additionally, the E(3) neural network (E3NN) is employed to predict the asymptotic parameters directly from the local geometry. The E3NN models demonstrate exceptional accuracy and transferability across all asymptotic parameters, achieving an R2 of 0.999 for both protein fragments and 20 amino acid dipeptide test sets. The long-range electrostatic and dispersion energies can be obtained using the E3NN-predicted parameters, with an error of 0.07 and 0.02 kcal/mol, respectively, when compared to symmetry-adapted perturbation theory (SAPT). Therefore, our force fields demonstrate the capability to accurately describe long-range interactions in proteins, paving the way for next-generation protein force fields.

Quantum-Informed Molecular Representation Learning Enhancing ADMET Property Prediction

We examined pretraining tasks leveraging abundant labeled data to effectively enhance molecular representation learning in downstream tasks, specifically emphasizing graph transformers to improve the prediction of ADMET properties. Our investigation revealed limitations in previous pretraining tasks and identified more meaningful training targets, ranging from 2D molecular descriptors to extensive quantum chemistry simulations. These data were seamlessly integrated into supervised pretraining tasks. The implementation of our pretraining strategy and multitask learning outperforms conventional methods, achieving state-of-the-art outcomes in 7 out of 22 ADMET tasks within the Therapeutics Data Commons by utilizing a shared encoder across all tasks. Our approach underscores the effectiveness of learning molecular representations and highlights the potential for scalability when leveraging extensive data sets, marking a significant advancement in this domain.

emle-engine: A Flexible Electrostatic Machine Learning Embedding Package for Multiscale Molecular Dynamics Simulations

We present in this work the emle-engine package (https://github.com/chemle/emle-engine)─the implementation of a new machine learning embedding scheme for hybrid machine learning potential/molecular-mechanics (ML/MM) dynamics simulations. The package is based on an embedding scheme that uses a physics-based model of the electronic density and induction with a handful of tunable parameters derived from in vacuo properties of the subsystem to be embedded. This scheme is completely independent of the in vacuo potential and requires only the positions of the atoms of the machine learning subsystem and the positions and partial charges of the molecular mechanics environment. These characteristics allow emle-engine to be employed in existing QM/MM software. We demonstrate that the implemented electrostatic machine learning embedding scheme (named EMLE) is stable in enhanced sampling molecular dynamics simulations. Through the calculation of free energy surfaces of alanine dipeptide in water with two different ML options for the in vacuo potential and three embedding models, we test the performance of EMLE. When compared to the reference DFT/MM surface, the EMLE embedding is clearly superior to the MM one based on fixed partial charges. The configurational dependence of the electronic density and the inclusion of the induction energy introduced by the EMLE model leads to a systematic reduction in the average error of the free energy surface when compared to MM embedding. By enabling the usage of EMLE embedding in practical ML/MM simulations, emle-engine will make it possible to accurately model systems and processes that feature significant variations in the charge distribution of the ML subsystem and/or the interacting environment.

Assessment of Embedding Schemes in a Hybrid Machine Learning/Classical Potentials (ML/MM) Approach

Machine learning (ML) methods have reached high accuracy levels for the prediction of in vacuo molecular properties. However, the simulation of large systems solely through ML methods (such as those based on neural network potentials) is still a challenge. In this context, one of the most promising frameworks for integrating ML schemes in the simulation of complex molecular systems are the so-called ML/MM methods. These multiscale approaches combine ML methods with classical force fields (MM), in the same spirit as the successful hybrid quantum mechanics-molecular mechanics methods (QM/MM). The key issue for such ML/MM methods is an adequate description of the coupling between the region of the system described by ML and the region described at the MM level. In the context of QM/MM schemes, the main ingredient of the interaction is electrostatic, and the state of the art is the so-called electrostatic-embedding. In this study, we analyze the quality of simpler mechanical embedding-based approaches, specifically focusing on their application within a ML/MM framework utilizing atomic partial charges derived in vacuo. Taking as reference electrostatic embedding calculations performed at a QM(DFT)/MM level, we explore different atomic charges schemes, as well as a polarization correction computed using atomic polarizabilites. Our benchmark data set comprises a set of about 80k small organic structures from the ANI-1x and ANI-2x databases, solvated in water. The results suggest that the minimal basis iterative stockholder (MBIS) atomic charges yield the best agreement with the reference coupling energy. Remarkable enhancements are achieved by including a simple polarization correction.

Ambiguities in Decomposing Molecular Polarizability into Atomic Charge Flow and Induced Dipole Contributions

The molecular dipole polarizability can be decomposed into components corresponding to the charge flow between atoms and changes in atomic dipole moments. Such decompositions are recognized to depend on how atoms are defined within a molecule, as, for example, by Hirshfeld, iterative Stockholder, or quantum topology partitioning of the electron density. For some of these, however, there are significant differences between the numerical results obtained by analytical response methods and finite field calculations. We show that this difference is due to analytical response methods accounting for (only) the change in electron density by a perturbation, while finite field methods may also include a component corresponding to a perturbation-dependent change in the definition of an atom within a molecule. For some atom-in-molecule definitions, such as the iterative Hirshfeld, iterative Stockholder, and quantum topology methods, the latter effect significantly increases the charge flow component. The decomposition of molecular polarizability into atomic charge flow and induced dipole components thus depends on whether the atom-in-molecule definition is taken to be perturbation-dependent.

On the sensitivity of computed partial charges toward basis set and (exchange-)correlation treatment

Partial charges are a central concept in general chemistry and chemical biology, yet dozens of different computational definitions exist. In prior work [Cho et al., ChemPhysChem 21, 688-696 (2020)], we showed that these can be reduced to at most three 'principal components of ionicity'. The present study addressed the dependence of computed partial charges q on 1-particle basis set and (for WFT methods) n -particle correlation treatment or (for DFT methods) exchange-correlation functional, for several representative partial charge definitions such as QTAIM, Hirshfeld, Hirshfeld-I, HLY (electrostatic), NPA, and GAPT. Our findings show that semi-empirical double hybrids can closely approach the CCSD(T) 'gold standard' for this property. In fact, owing to an error compensation in MP2, CCSD partial charges are further away from CCSD(T) than is MP2. The nonlocal correlation is important, especially when there is a substantial amount of nonlocal exchange. Employing range separation proves to be "mostly" not advantageous, while global hybrids perform optimally for 20%-30% Hartree-Fock exchange across all charge types. Basis set convergence analysis shows that an augmented triple-zeta heavy-aug-cc-pV(T+d)Z basis set or a partially augmented jun-cc-pV(T+d)Z basis set is sufficient for Hirshfeld, Hirshfeld-I, HLY, and GAPT charges. In contrast, QTAIM and NPA display slower basis set convergence. It is noteworthy that for both NPA and QTAIM, HF exhibits markedly slower basis set convergence than the correlation components of MP2 and CCSD. Triples corrections in CCSD(T), denoted as CCSD(T)-CCSD, exhibit even faster basis set convergence.

The tale of HORTON: Lessons learned in a decade of scientific software development

HORTON is a free and open-source electronic-structure package written primarily in Python 3 with some underlying C++ components. While HORTON's development has been mainly directed by the research interests of its leading contributing groups, it is designed to be easily modified, extended, and used by other developers of quantum chemistry methods or post-processing techniques. Most importantly, HORTON adheres to modern principles of software development, including modularity, readability, flexibility, comprehensive documentation, automatic testing, version control, and quality-assurance protocols. This article explains how the principles and structure of HORTON have evolved since we started developing it more than a decade ago. We review the features and functionality of the latest HORTON release (version 2.3) and discuss how HORTON is evolving to support electronic structure theory research for the next decade.

Nonbonded Force Field Parameters from MBIS Partitioning of the Molecular Electron Density Improve Binding Affinity Predictions of the T4-Lysozyme Double Mutant

The use of computer simulation for binding affinity prediction is growing in drug discovery. However, its wider use is constrained by the accuracy of the free energy calculations. The key sources of error are the force fields used to depict molecular interactions and insufficient sampling of the configurational space. To improve the quality of the force field, we developed a Python-based computational workflow. The workflow described here uses the minimal basis iterative stockholder (MBIS) method to determine atomic charges and Lennard-Jones parameters from the polarized molecular density. This is done by performing electronic structure calculations on various configurations of the ligand when it is both bound and unbound. In addition, we validated a simulation procedure that accounts for the protein and ligand degrees of freedom to precisely calculate binding free energies. This was achieved by comparing the self-adjusted mixture sampling and nonequilibrium thermodynamic integration methods using various protein and ligand conformations. The accuracy of predicting binding affinity is improved by using MBIS-derived force field parameters and a validated simulation procedure. This improvement surpasses the chemical precision for the eight aromatic ligands, reaching a root-mean-square error of 0.7 kcal/mol.

Force-field-enhanced neural network interactions: from local equivariant embedding to atom-in-molecule properties and long-range effects

We introduce FENNIX (Force-Field-Enhanced Neural Network InteraXions), a hybrid approach between machine-learning and force-fields. We leverage state-of-the-art equivariant neural networks to predict local energy contributions and multiple atom-in-molecule properties that are then used as geometry-dependent parameters for physically-motivated energy terms which account for long-range electrostatics and dispersion. Using high-accuracy ab initio data (small organic molecules/dimers), we trained a first version of the model. Exhibiting accurate gas-phase energy predictions, FENNIX is transferable to the condensed phase. It is able to produce stable Molecular Dynamics simulations, including nuclear quantum effects, for water predicting accurate liquid properties. The extrapolating power of the hybrid physically-driven machine learning FENNIX approach is exemplified by computing: (i) the solvated alanine dipeptide free energy landscape; (ii) the reactive dissociation of small molecules.

Hybrid classical/machine-learning force fields for the accurate description of molecular condensed-phase systems

Electronic structure methods offer in principle accurate predictions of molecular properties, however, their applicability is limited by computational costs. Empirical methods are cheaper, but come with inherent approximations and are dependent on the quality and quantity of training data. The rise of machine learning (ML) force fields (FFs) exacerbates limitations related to training data even further, especially for condensed-phase systems for which the generation of large and high-quality training datasets is difficult. Here, we propose a hybrid ML/classical FF model that is parametrized exclusively on high-quality ab initio data of dimers and monomers in vacuum but is transferable to condensed-phase systems. The proposed hybrid model combines our previous ML-parametrized classical model with ML corrections for situations where classical approximations break down, thus combining the robustness and efficiency of classical FFs with the flexibility of ML. Extensive validation on benchmarking datasets and experimental condensed-phase data, including organic liquids and small-molecule crystal structures, showcases how the proposed approach may promote FF development and unlock the full potential of classical FFs.

DASH: Dynamic Attention-Based Substructure Hierarchy for Partial Charge Assignment

We present a robust and computationally efficient approach for assigning partial charges of atoms in molecules. The method is based on a hierarchical tree constructed from attention values extracted from a graph neural network (GNN), which was trained to predict atomic partial charges from accurate quantum-mechanical (QM) calculations. The resulting dynamic attention-based substructure hierarchy (DASH) approach provides fast assignment of partial charges with the same accuracy as the GNN itself, is software-independent, and can easily be integrated in existing parametrization pipelines, as shown for the Open force field (OpenFF). The implementation of the DASH workflow, the final DASH tree, and the training set are available as open source/open data from public repositories.

An information-theoretic approach to basis-set fitting of electron densities and other non-negative functions

The numerical ill-conditioning associated with approximating an electron density with a convex sum of Gaussian or Slater-type functions is overcome by using the (extended) Kullback-Leibler divergence to measure the deviation between the target and approximate density. The optimized densities are non-negative and normalized, and they are accurate enough to be used in applications related to molecular similarity, the topology of the electron density, and numerical molecular integration. This robust, efficient, and general approach can be used to fit any non-negative normalized functions (e.g., the kinetic energy density and molecular electron density) to a convex sum of non-negative basis functions. We present a fixed-point iteration method for optimizing the Kullback-Leibler divergence and compare it to conventional gradient-based optimization methods. These algorithms are released through the free and open-source BFit package, which also includes a L2-norm squared optimization routine applicable to any square-integrable scalar function.

Assessment of an Electrostatic Energy-Based Charge Model for Modeling the Electrostatic Interactions in Water Solvent

The protein force field based on the restrained electrostatic potential (RESP) charges has limitations in accurately describing hydrogen bonding interactions in proteins. To address this issue, we propose an alternative approach called the electrostatic energy-based charges (EEC) model, which shows improved performance in describing electrostatic interactions (EIs) of hydrogen bonds in proteins. In this study, we further investigate the performance of the EEC model in modeling EIs in water solvent. Our findings demonstrate that the fixed EEC model can effectively reproduce the quantum mechanics/molecular mechanics (QM/MM)-calculated EIs between a water molecule and various water solvent environments. However, to achieve the same level of computational accuracy, the electrostatic potential (ESP) charge model needs to fluctuate according to the electrostatic environment. Our analysis indicates that the requirement for charge adjustments depends on the specific mathematical and physical representation of EIs as a function of the environment for deriving charges. By comparing with widely used empirical water models calibrated to reproduce experimental properties, we confirm that the performance of the EEC model in reproducing QM/MM EIs is similar to that of general purpose TIP4P-like water models such as TIP4P-Ew and TIP4P/2005. When comparing the computed 10,000 distinct EI values within the range of -40 to 0 kcal/mol with the QM/MM results calculated at the MP2/aug-cc-pVQZ/TIP3P level, we noticed that the mean unsigned error (MUE) for the EEC model is merely 0.487 kcal/mol, which is remarkably similar to the MUE values of the TIP4P-Ew (0.63 kcal/mol) and TIP4P/2005 (0.579 kcal/mol) models. However, both the RESP method and the TIP3P model exhibit a tendency to overestimate the EIs, as evidenced by their higher MUE values of 1.761 and 1.293 kcal/mol, respectively. EEC-based molecular dynamics simulations have demonstrated that, when combined with appropriate van der Waals parameters, the EEC model can closely reproduce oxygen-oxygen radial distribution function and density of water, showing a remarkable similarity to the well-established TIP4P-like empirical water models. Our results demonstrate that the EEC model has the potential to build force fields with comparable accuracy to more sophisticated empirical TIP4P-like water models.

Quantifying Intramolecular Basis Set Superposition Errors

We show that medium-sized Gaussian basis sets lead to significant intramolecular basis set superposition errors at Hartree-Fock and density functional levels of theory, with artificial stabilization of compact over extended conformations for a 186 atom deca-peptide. Errors of ∼80 and ∼10 kJ/mol are observed, with polarized double zeta and polarized triple zeta quality basis sets, respectively. Two different procedures for taking the basis set superposition error into account are tested. While both reduce the error, it appears that polarized quadruple zeta basis sets are required to reduce the error below a few kJ/mol. Alternatively, the basis set superposition error can be eliminated using multiresolution methods based on Multiwavelets.

Application of message passing neural networks for molecular property prediction

Accurate molecular property prediction, as one of the classical cheminformatics topics, plays a prominent role in the fields of computer-aided drug design. For instance, property prediction models can be used to quickly screen large molecular libraries to find lead compounds. Message-passing neural networks (MPNNs), a sub-class of Graph neural networks (GNNs), have recently been demonstrated to outperform other deep learning methods on a variety of tasks, including the prediction of molecular characteristics. In this survey, we provide a brief review of the MPNN models and their applications on molecular property prediction.

Unifying Charge-Flow Polarization Models

We show that several models where electric polarization in molecular systems is modeled by charge-flow between atoms can all be considered as different manifestations of a general underlying mathematical structure. The models can be classified according to whether they employ atomic or bond parameters and whether they employ atom/bond hardness or softness. We show that an ab initio calculated charge response kernel can be considered as the inverse screened Coulombic matrix projected onto the zero-charge subspace, and this may provide a method for deriving charge screening functions to be used in force fields. The analysis suggests that some models contain redundancies, and we argue that a parameterization of charge-flow models in terms of bond softness is preferable as it depends on local quantities and decay to zero upon bond dissociation, while bond hardness depends on global quantities and increases toward infinity upon bond dissociation.

On the role of history-dependent adsorbate distribution and metastable states in switchable mesoporous metal-organic frameworks

A unique feature of metal-organic frameworks (MOFs) in contrast to rigid nanoporous materials is their structural switchabilty offering a wide range of functionality for sustainable energy storage, separation and sensing applications. This has initiated a series of experimental and theoretical studies predominantly aiming at understanding the thermodynamic conditions to transform and release gas, but the nature of sorption-induced switching transitions remains poorly understood. Here we report experimental evidence for fluid metastability and history-dependent states during sorption triggering the structural change of the framework and leading to the counterintuitive phenomenon of negative gas adsorption (NGA) in flexible MOFs. Preparation of two isoreticular MOFs differing by structural flexibility and performing direct in situ diffusion studies aided by in situ X-ray diffraction, scanning electron microscopy and computational modelling, allowed assessment of n-butane molecular dynamics, phase state, and the framework response to obtain a microscopic picture for each step of the sorption process.

The primary photolysis of aqueous carbonate di-anions

We study the primary photolysis dynamics of aqueous carbonate, CO₃^2-(aq), and hydrogen carbonate, HCO₃^-(aq), when they are excited at λ = 200 nm. The photolysis is recorded with sub-picosecond time resolution using UV pump-Vis probe and UV pump-IR probe transient absorption spectroscopy and interpreted with the aid of density functional theory calculations. When CO₃^2- is excited via single photon absorption at λ = 200 nm, Φ(t = 20 ps) = 82 ± 5% of the excited di-anions either detach an electron or dissociate. The electron detachment takes place from the excited state in t < 1 ps and forms ground state CO₃˙^- and e_aq^-. Dissociation occurs from both the electronic ground and excited states of CO₃^2-. Dissociation from the CO₃^2- excited state is assisted by water molecules and forms CO₂˙^-, OH˙ and OH^-. The dissociation occurs both directly from the Franck-Condon region in t < 1 ps and indirectly with a time constant of τ = 13.9 ± 0.5 ps as the excited state relaxes. Dissociation of vibrationally excited CO₃^2- molecules in the electronic ground state is also assisted by water molecules and forms CO₂ and two OH^- anions. The dissociation and subsequent vibrational relaxation of CO₂ occur with a time constant of τ = 10.2 ± 0.5 ps. The residual 1 - Φ(t = 20 ps) = 18 ± 5% of the excited CO₃^2- di-anions return by internal conversion to the equilibrated CO₃^2- ground state with a time constant of τ = 4.0 ± 0.4 ps. The extinction coefficient of aqueous hydrogen carbonate HCO₃^-(aq) at λ = 200 nm is an order of magnitude smaller than that of carbonate, so even though the hydrogen carbonate anions dominate the carbonate di-anions in the hydrogen carbonate solution, the primary photolysis of hydrogen carbonate is obscured by the photo-products of carbonate. Hence, we are unable to assess the primary photolysis of hydrogen carbonate. However, the weak one-photon absorption facilitates two-photon ionization of water, which forms hydronium, H₃O⁺, cations. The sudden increase in the acidity induced by two-photon ionization protonates the ground state hydrogen carbonate molecules, thus offering a rare spectroscopic glimpse of aqueous carbonic acid.

Sensitivity Analysis for ReaxFF Reparametrization Using the Hilbert-Schmidt Independence Criterion

We apply a global sensitivity method, the Hilbert-Schmidt independence criterion (HSIC), to the reparametrization of a Zn/S/H ReaxFF force field to identify the most appropriate parameters for reparametrization. Parameter selection remains a challenge in this context, as high-dimensional optimizations are prone to overfitting and take a long time but selecting too few parameters leads to poor-quality force fields. We show that the HSIC correctly and quickly identifies the most sensitive parameters and that optimizations done using a small number of sensitive parameters outperform those done using a higher-dimensional reasonable-user parameter selection. Optimizations using only sensitive parameters (1) converge faster, (2) have loss values comparable to those found with the naive selection, (3) have similar accuracy in validation tests, and (4) do not suffer from problems of overfitting. We demonstrate that an HSIC global sensitivity is a cheap optimization preprocessing step that has both qualitative and quantitative benefits which can substantially simplify and speed up ReaxFF reparametrizations.

Atomic charges in molecules defined by molecular real space partition into atomic subspaces

Atomic charge (AC), which is the charge distribution of a molecule, is an important property that is closely associated with structures, reactivities, and intra- and inter-molecular interactions among molecules. Several theoretical models or methods can be used to obtain the magnitudes of AC with different characteristics. These models can be classified into fuzzy-atoms models and models partitioning a molecule into individual atoms with sharp boundaries. The first category includes Mulliken, natural population analysis (NPA), Hirshfeld, Merz-Kollman-Singh (MK), CHELPG, the electronegativity equalization method (EEM), the atom-bond electronegativity equalization method (ABEEM), and atomic polar tensor (APT). The second category is derived from quantum chemical topology (QCT) and includes the quantum theory of atoms in molecules (QTAIM) and QCT analysis based on the potential acting on one electron in a molecule (PAEMQCT). Herein, after giving a bird's-eye view of the population methods of the first category, we specifically describe some features of the second category. We only present the basic framework of QCT for obtaining ACs from QTAIM and PAEMQCT and show their important characteristics. QCT establishes the basis of the following chemical concept: a molecule is spatially partitioned into individual atoms with sharp boundaries. The ACs from QTAIM are close to the atomic valence in chemistry, and ACs from PAEMQCT may be practically suitable for modeling intra- and inter-molecular interactions.

What is the Optimal Dipole Moment for Nonpolarizable Models of Liquids?

In classical nonpolarizable models, electrostatic interactions are usually described by assigning fixed partial charges to interaction sites. Despite the multitude of methods and theories proposed over the years for partial charge assignment, a fundamental question remains─what is the correct degree of polarization that a fixed-charge model should possess to provide the best balance of interactions (including induction effects) and yield the best description of the potential energy surface of a liquid phase? We address this question by approaching it from two separate and independent viewpoints: the QUantum mechanical BEspoke (QUBE) approach, which assigns bespoke force field parameters for individual molecules from ab initio calculations with minimal empirical fitting, and the Polarization-Consistent Approach (PolCA) force field, based on empirical fitting of force field parameters with an emphasis on transferability by rigorously accounting for polarization effects in the parameterization process. We show that the two approaches yield consistent answers to the above question, namely, that the dipole moment of the model should be approximately halfway between those of the gas and the liquid phase. Crucially, however, the reference liquid-phase dipole needs to be estimated using methods that explicitly consider both mean-field and local contributions to polarization. In particular, continuum dielectric models are inadequate for this purpose because they cannot account for local effects and therefore significantly underestimate the degree of polarization of the molecule. These observations have profound consequences for the development, validation, and testing of nonpolarizable models.

Electrostatic Embedding of Machine Learning Potentials

This work presents a variant of an electrostatic embedding scheme that allows the embedding of arbitrary machine learned potentials trained on molecular systems in vacuo. The scheme is based on physically motivated models of electronic density and polarizability, resulting in a generic model without relying on an exhaustive training set. The scheme only requires in vacuo single point QM calculations to provide training densities and molecular dipolar polarizabilities. As an example, the scheme is applied to create an embedding model for the QM7 data set using Gaussian Process Regression with only 445 reference atomic environments. The model was tested on the SARS-CoV-2 protease complex with PF-00835231, resulting in a predicted embedding energy RMSE of 2 kcal/mol, compared to explicit DFT/MM calculations.

Generalized Many-Body Dispersion Correction through Random-Phase Approximation for Chemically Accurate Density Functional Theory

We extend our recently proposed Deep Learning-aided many-body dispersion (DNN-MBD) model to quadrupole polarizability (Q) terms using a generalized Random Phase Approximation (RPA) formalism, thus enabling the inclusion of van der Waals contributions beyond dipole. The resulting DNN-MBDQ model only relies on ab initio-derived quantities as the introduced quadrupole polarizabilities are recursively retrieved from dipole ones, in turn modeled via the Tkatchenko-Scheffler method. A transferable and efficient deep-neuronal network (DNN) provides atom-in-molecule volumes, while a single range-separation parameter is used to couple the model to Density Functional Theory (DFT). Since it can be computed at a negligible cost, the DNN-MBDQ approach can be coupled with DFT functionals, such as PBE, PBE0, and B86bPBE (dispersionless). The DNN-MBQ-corrected functionals reach chemical accuracy while exhibiting lower errors compared to their dipole-only counterparts.

Regularized by Physics: Graph Neural Network Parametrized Potentials for the Description of Intermolecular Interactions

Simulations of molecular systems using electronic structure methods are still not feasible for many systems of biological importance. As a result, empirical methods such as force fields (FF) have become an established tool for the simulation of large and complex molecular systems. The parametrization of FF is, however, time-consuming and has traditionally been based on experimental data. Recent years have therefore seen increasing efforts to automatize FF parametrization or to replace FF with machine-learning (ML) based potentials. Here, we propose an alternative strategy to parametrize FF, which makes use of ML and gradient-descent based optimization while retaining a functional form founded in physics. Using a predefined functional form is shown to enable interpretability, robustness, and efficient simulations of large systems over long time scales. To demonstrate the strength of the proposed method, a fixed-charge and a polarizable model are trained on ab initio potential-energy surfaces. Given only information about the constituting elements, the molecular topology, and reference potential energies, the models successfully learn to assign atom types and corresponding FF parameters from scratch. The resulting models and parameters are validated on a wide range of experimentally and computationally derived properties of systems including dimers, pure liquids, and molecular crystals.

gmak: A Parameter-Space Mapping Strategy for Force-Field Calibration

In the context of classical molecular simulations, the accuracy of a force field is highly influenced by the values of the relevant simulation parameters. In this work, a parameter-space mapping (PSM) workflow is proposed to aid in the calibration of force-field parameters, based mainly on the following features: (i) regular-grid discretization of the search space; (ii) partial sampling of the search-space grid; (iii) training of surrogate models to predict the estimates of the target properties for nonsampled parameter sets; (iv) post hoc interpretation of the results in terms of multiobjective optimization concepts; (v) attenuation of statistical errors achieved via empiric extension of the duration of the simulations; (vi) iterative search-space translation according to a user-defined scalar objective function that measures the accuracy of the force field (e.g., the weighted root-mean-square deviation of the target properties relative to the reference data). This combination of features results in a hybrid of a single- and a multiobjective optimization strategy, allowing for the approximate determination of both a local minimum of the chosen objective function and its neighboring Pareto efficient points. The PSM workflow is implemented in the extensible Python program gmak, which is made available in the Git repository at http://github.com/mssm-labmmol/gmak. Using this implementation, the PSM workflow was tested in a proof-of-concept fashion in the recalibration of the Lennard-Jones parameters of the 3-point Optimal Point Charge (OPC3) water model for compatibility with the GROMOS treatment of nonbonded interactions. The recalibrated model reproduces typical pure-liquid properties with an accuracy similar to the original OPC3 model and represents a significant improvement relative to the Simple Point Charge (SPC) model, which is the official recommendation for simulations using GROMOS force fields.

Comparison of the United- and All-Atom Representations of (Halo)alkanes Based on Two Condensed-Phase Force Fields Optimized against the Same Experimental Data Set

The level of accuracy that can be achieved by a force field is influenced by choices made in the interaction-function representation and in the relevant simulation parameters. These choices, referred to here as functional-form variants (FFVs), include for example the model resolution, the charge-derivation procedure, the van der Waals combination rules, the cutoff distance, and the treatment of the long-range interactions. Ideally, assessing the effect of a given FFV on the intrinsic accuracy of the force-field representation requires that only the specific FFV is changed and that this change is performed at an optimal level of parametrization, a requirement that may prove extremely challenging to achieve in practice. Here, we present a first attempt at such a comparison for one specific FFV, namely the choice of a united-atom (UA) versus an all-atom (AA) resolution in a force field for saturated acyclic (halo)alkanes. Two force-field versions (UA vs AA) are optimized in an automated way using the CombiFF approach against 961 experimental values for the pure-liquid densities ρ_liq and vaporization enthalpies ΔH_vap of 591 compounds. For the AA force field, the torsional and third-neighbor Lennard-Jones parameters are also refined based on quantum-mechanical rotational-energy profiles. The comparison between the UA and AA resolutions is also extended to properties that have not been included as parameterization targets, namely the surface-tension coefficient γ, the isothermal compressibility κ_T, the isobaric thermal-expansion coefficient α_P, the isobaric heat capacity c_P, the static relative dielectric permittivity ϵ, the self-diffusion coefficient D, the shear viscosity η, the hydration free energy ΔG_wat, and the free energy of solvation ΔG_che in cyclohexane. For the target properties ρ_liq and ΔH_vap, the UA and AA resolutions reach very similar levels of accuracy after optimization. For the nine other properties, the AA representation leads to more accurate results in terms of η; comparably accurate results in terms of γ, κ_T, α_P, ϵ, D, and ΔG_che; and less accurate results in terms of c_P and ΔG_wat. This work also represents a first step toward the calibration of a GROMOS-compatible force field at the AA resolution.

A new framework for frequency-dependent polarizable force fields

A frequency-dependent extension of the polarizable force field "Atom-Condensed Kohn-Sham density functional theory approximated to the second-order" (ACKS2) [Verstraelen et al., J. Chem. Phys. 141, 194114 (2014)] is proposed, referred to as ACKS2ω. The method enables theoretical predictions of dynamical response properties of finite systems after partitioning of the frequency-dependent molecular response function. Parameters in this model are computed simply as expectation values of an electronic wavefunction, and the hardness matrix is entirely reused from ACKS2 as an adiabatic approximation is used. A numerical validation shows that accurate models can already be obtained with atomic monopoles and dipoles. Absorption spectra of 42 organic and inorganic molecular monomers are evaluated using ACKS2ω, and our results agree well with the time-dependent DFT calculations. Also for the calculation of C₆ dispersion coefficients, ACKS2ω closely reproduces its TDDFT reference. When parameters for ACKS2ω are derived from a PBE/aug-cc-pVDZ ground state, it reproduces experimental values for 903 organic and inorganic intermolecular pairs with an MAPE of 3.84%. Our results confirm that ACKS2ω offers a solid connection between the quantum-mechanical description of frequency-dependent response and computationally efficient force-field models.

Non-covalent interactions from a Quantum Chemical Topology perspective

About half a century after its little-known beginnings, the quantum topological approach called QTAIM has grown into a widespread, but still not mainstream, methodology of interpretational quantum chemistry. Although often confused in textbooks with yet another population analysis, be it perhaps an elegant but somewhat esoteric one, QTAIM has been enriched with about a dozen other research areas sharing its main mathematical language, such as Interacting Quantum Atoms (IQA) or Electron Localisation Function (ELF), to form an overarching approach called Quantum Chemical Topology (QCT). Instead of reviewing the latter's role in understanding non-covalent interactions, we propose a number of ideas emerging from the full consequences of the space-filling nature of topological atoms, and discuss how they (will) impact on interatomic interactions, including non-covalent ones. The architecture of a force field called FFLUX, which is based on these ideas, is outlined. A new method called Relative Energy Gradient (REG) is put forward, which is able, by computation, to detect which fragments of a given molecular assembly govern the energetic behaviour of this whole assembly. This method can offer insight into the typical balance of competing atomic energies both in covalent and non-covalent case studies. A brief discussion on so-called bond critical points is given, highlighting concerns about their meaning, mainly in the arena of non-covalent interactions.

Nonbonded Force Field Parameters from Minimal Basis Iterative Stockholder Partitioning of the Molecular Electron Density Improve CB7 Host-Guest Affinity Predictions

Binding affinity prediction by means of computer simulation has been increasingly incorporated in drug discovery projects. Its wide application, however, is limited by the prediction accuracy of the free energy calculations. The main error sources are force fields used to describe molecular interactions and incomplete sampling of the configurational space. Organic host-guest systems have been used to address force field quality because they share similar interactions found in ligands and receptors, and their rigidity facilitates configurational sampling. Here, we test the binding free energy prediction accuracy for 14 guests with an aromatic or adamantane core and the CB7 host using molecular electron density derived nonbonded force field parameters. We developed a computational workflow written in Python to derive atomic charges and Lennard-Jones parameters with the Minimal Basis Iterative Stockholder method using the polarized electron density of several configurations of each guest in the bound and unbound states. The resulting nonbonded force field parameters improve binding affinity prediction, especially for guests with an adamantane core in which repulsive exchange and dispersion interactions to the host dominate.

Exploration and validation of force field design protocols through QM-to-MM mapping

The scale of the parameter optimisation problem in traditional molecular mechanics force field construction means that design of a new force field is a long process, and sub-optimal choices made in the early stages can persist for many generations. We hypothesise that careful use of quantum mechanics to inform molecular mechanics parameter derivation (QM-to-MM mapping) should be used to significantly reduce the number of parameters that require fitting to experiment and increase the pace of force field development. Here, we design and train a collection of 15 new protocols for small, organic molecule force field derivation, and test their accuracy against experimental liquid properties. Our best performing model has only seven fitting parameters, yet achieves mean unsigned errors of just 0.031 g cm^-3 and 0.69 kcal mol^-1 in liquid densities and heats of vaporisation, compared to experiment. The software required to derive the designed force fields is freely available at https://github.com/qubekit/QUBEKit.

Combined Deep Learning and Classical Potential Approach for Modeling Diffusion in UiO-66

Modeling of diffusion of adsorbates through porous materials with atomistic molecular dynamics (MD) can be a challenging task if the flexibility of the adsorbent needs to be included. This is because potentials need to be developed that accurately account for the motion of the adsorbent in response to the presence of adsorbate molecules. In this work, we show that it is possible to use accurate machine learning atomistic potentials for metal-organic frameworks in concert with classical potentials for adsorbates to accurately compute diffusivities though a hybrid potential approach. As a proof-of-concept, we have developed an accurate deep learning potential (DP) for UiO-66, a metal-organic framework, and used this DP to perform hybrid potential simulations, modeling diffusion of neon and xenon through the crystal. The adsorbate-adsorbate interactions were modeled with Lennard-Jones (LJ) potentials, the adsorbent-adsorbent interactions were described by the DP, and the adsorbent-adsorbate interactions used LJ cross-interactions. Thus, our hybrid potential allows for adsorbent-adsorbate interactions with classical potentials but models the response of the adsorbent to the presence of the adsorbate through near-DFT accuracy DPs. This hybrid approach does not require refitting the DP for new adsorbates. We calculated self-diffusion coefficients for Ne in UiO-66 from DFT-MD, our hybrid DP/LJ approach, and from two different classical potentials for UiO-66. Our DP/LJ results are in excellent agreement with DFT-MD. We modeled diffusion of Xe in UiO-66 with DP/LJ and a classical potential. Diffusion of Xe in UiO-66 is about a factor of 30 slower than that of Ne, so it is not computationally feasible to compute Xe diffusion with DFT-MD. Our hybrid DP-classical potential approach can be applied to other MOFs and other adsorbates, making it possible to use an accurate DP generated from DFT simulations of an empty adsorbent in concert with existing classical potentials for adsorbates to model adsorption and diffusion within the porous material, including adsorbate-induced changes to the framework.

Constrained iterative Hirshfeld charges: A variational approach

We develop a variational procedure for the iterative Hirshfeld (HI) partitioning scheme. The main practical advantage of having a variational framework is that it provides a formal and straightforward approach for imposing constraints (e.g., fixed charges on certain atoms or molecular fragments) when computing HI atoms and their properties. Unlike many other variants of the Hirshfeld partitioning scheme, HI charges do not arise naturally from the information-theoretic framework, but only as a reverse-engineered construction of the objective function. However, the procedure we use is quite general and could be applied to other problems as well. We also prove that there is always at least one solution to the HI equations, but we could not prove that its self-consistent equations would always converge for any given initial pro-atom charges. Our numerical assessment of the constrained iterative Hirshfeld method shows that it satisfies many desirable traits of atoms in molecules and has the potential to surpass existing approaches for adding constraints when computing atomic properties.

Accurate Deep Learning-Aided Density-Free Strategy for Many-Body Dispersion-Corrected Density Functional Theory

Using a deep neuronal network (DNN) model trained on the large ANI-1 data set of small organic molecules, we propose a transferable density-free many-body dispersion (DNN-MBD) model. The DNN strategy bypasses the explicit Hirshfeld partitioning of the Kohn-Sham electron density required by MBD models to obtain the atom-in-molecules volumes used by the Tkatchenko-Scheffler polarizability rescaling. The resulting DNN-MBD model is trained with minimal basis iterative Stockholder atomic volumes and, coupled to density functional theory (DFT), exhibits comparable (if not greater) accuracy to other approaches based on different partitioning schemes. Implemented in the Tinker-HP package, the DNN-MBD model decreases the overall computational cost compared to MBD models where the explicit density partitioning is performed. Its coupling with the recently introduced Stochastic formulation of the MBD equations (J. Chem. Theory Comput. 2022, 18 (3), 1633-1645) enables large routine dispersion-corrected DFT calculations at preserved accuracy. Furthermore, the DNN electron density-free features extend the MBD model's applicability beyond electronic structure theory within methodologies such as force fields and neural networks.

Multi-center decomposition of molecular densities: a mathematical perspective

The aim of this article is to analyze from a mathematical perspective some existing schemes to partition a molecular density into several atomic contributions, with a specific focus on Iterative Stockholder Atom (ISA) methods. We provide a unified mathematical framework to describe the latter family of methods and propose a new scheme, named L-ISA (for linear approximation of ISA), which generalizes the so-called Additive Variational Hirshfeld method. We prove several important mathematical properties of the ISA and L-ISA minimization problems and show that the so-called ISA algorithms can be viewed as alternating minimization schemes, which in turn enables us to obtain new convergence results for these numerical methods. Specific mathematical properties of the ISA decomposition for diatomic systems are also presented. Different schemes are numerically compared on different molecules and we discuss the advantages and drawbacks of each approach. Numerical results on diatomic systems illustrate the proven mathematical properties.

Learning Atomic Multipoles: Prediction of the Electrostatic Potential with Equivariant Graph Neural Networks

The accurate description of electrostatic interactions remains a challenging problem for classical potential-energy functions. The commonly used fixed partial-charge approximation fails to reproduce the electrostatic potential at short range due to its insensitivity to conformational changes and anisotropic effects. At the same time, possibly more accurate machine-learned (ML) potentials struggle with the long-range behavior due to their inherent locality ansatz. Employing a multipole expansion offers in principle an exact treatment of the electrostatic potential such that the long-range and short-range electrostatic interactions can be treated simultaneously with high accuracy. However, such an expansion requires the calculation of the electron density using computationally expensive quantum-mechanical (QM) methods. Here, we introduce an equivariant graph neural network (GNN) to address this issue. The proposed model predicts atomic multipoles up to the quadrupole, circumventing the need for expensive QM computations. By using an equivariant architecture, the model enforces the correct symmetry by design without relying on local reference frames. The GNN reproduces the electrostatic potential of various systems with high fidelity. Possible uses for such an approach include the separate treatment of long-range interactions in ML potentials, the analysis of electrostatic potential surfaces, and static multipoles in polarizable force fields.

Using atomic charges to model molecular polarization

We review different models for introducing electric polarization in force fields, with special focus on methods where polarization is modelled at the atomic charge level. While electric polarization has been included in several force fields, the common approach has been to focus on atomic dipole polarizability. Several approaches allow modelling electric polarization by using charge-flow between charge sites instead, but this has been less exploited, despite that atomic charges and charge-flow is expected to be more important than atomic dipoles and dipole polarizability. A number of challenges are required to be solved for charge-flow models to be incorporated into polarizable force fields, for example how to parameterize the models and how to make them computational efficient.

NNAIMQ: A neural network model for predicting QTAIM charges

Atomic charges provide crucial information about the electronic structure of a molecular system. Among the different definitions of these descriptors, the one proposed by the Quantum Theory of Atoms in Molecules (QTAIM) is particularly attractive given its invariance against orbital transformations although the computational cost associated with their calculation limits its applicability. Given that Machine Learning (ML) techniques have been shown to accelerate orders of magnitude the computation of a number of quantum mechanical observables, in this work, we take advantage of ML knowledge to develop an intuitive and fast neural network model (NNAIMQ) for the computation of QTAIM charges for C, H, O, and N atoms with high accuracy. Our model has been trained and tested using data from quantum chemical calculations in more than 45 000 molecular environments of the near-equilibrium CHON chemical space. The reliability and performance of NNAIMQ have been analyzed in a variety of scenarios, from equilibrium geometries to molecular dynamics simulations. Altogether, NNAIMQ yields remarkably small prediction errors, well below the 0.03 electron limit in the general case, while accelerating the calculation of QTAIM charges by several orders of magnitude.

Apples to apples comparison of standardized to unstandardized principal component analysis of methods that assign partial atomic charges in molecules

Articles by Cho et al. (ChemPhysChem, 2020, 21, 688–696) and Manz (RSC Adv., 2020, 10, 44121–44148) performed unstandardized and standardized, respectively, principal component analysis (PCA) to study atomic charge assignment methods for molecular systems. Both articles used subsets of atomic charges computed by Cho et al.; however, the data subsets employed were not strictly identical. Herein, an element by element analysis of this dataset is first performed to compare the spread of charge values across individual chemical elements and charge assignment methods. This reveals an underlying problem with the reported Becke partial atomic charges in this dataset. Due to their unphysical values, these Becke charges were not included in the subsequent PCA. Standardized and unstandardized PCA are performed across two datasets: (i) 19 charge assignment methods having a complete basis set limit and (ii) all 25 charge assignment methods (excluding Becke) for which Cho et al. computed atomic charges. The dataset contained ∼2000 molecules having a total of 29 907 atoms in materials. The following five methods (listed here in alphabetical order) showed the greatest correlation to the first principal component in standardized and unstandardized PCA: DDEC6, Hirshfeld-I, ISA, MBIS, and MBSBickelhaupt (note: MBSBickelhaupt does not appear in the 19 methods dataset). For standardized PCA, the DDEC6 method ranked first followed closely by MBIS. For unstandardized PCA, Hirshfeld-I (19 methods) or MBSBickelhaupt (25 methods) ranked first followed by DDEC6 in second place (both 19 and 25 methods).

Standardized and unstandardized principal component analyses are directly compared for a dataset of ∼2000 molecules across various charge assignment methods.