Accurate Force Field Research Articles

Anomalous Water Penetration in Al3+ Dissolution.

The physicochemical characterization of trivalent ions is limited due to a lack of accurate force fields. By leveraging the latest machine learning force field to model aqueous AlCl3, we discover that upon dissolution of Al3+, water molecules beyond the second hydration shell are involved in the hydration process. A combination of scissoring of coordinating water is followed by synchronized secondary motion of water in the second solvation shell due to hydrogen bonding. Consequently, the water beyond the second solvation penetrates through the second solvation shell and coordinates to the Al3+. Our study reveals a novel microscopic understanding of solvation dynamics for the trivalent ion.

Alchemical Transformations and Beyond: Recent Advances and Real-World Applications of Free Energy Calculations in Drug Discovery.

Computational methods constitute efficient strategies for screening and optimizing potential drug molecules. A critical factor in this process is the binding affinity between candidate molecules and targets, quantified as binding free energy. Among various estimation methods, alchemical transformation methods stand out for their theoretical rigor. Despite challenges in force field accuracy and sampling efficiency, advancements in algorithms, software, and hardware have increased the application of free energy perturbation (FEP) calculations in the pharmaceutical industry. Here, we review the practical applications of FEP in drug discovery projects since 2018, covering both ligand-centric and residue-centric transformations. We show that relative binding free energy calculations have steadily achieved chemical accuracy in real-world applications. In addition, we discuss alternative physics-based simulation methods and the incorporation of deep learning into free energy calculations.

Exploiting Metal-Organic Frameworks for Vinylidene Fluoride Adsorption: From Force Field Development, Computational Screening to Machine Learning.

Metal-organic frameworks (MOFs) represent a distinctive class of nanoporous materials with considerable potential across a wide range of applications. Recently, a handful of MOFs has been explored for the storage of environmentally hazardous fluorinated gases (Keasler et al. Science 2023, 381, 1455), yet the potential of over 100,000 MOFs for this specific application has not been thoroughly investigated, particularly due to the absence of an established force field. In this study, we develop an accurate force field for nonaversive hydrofluorocarbon vinylidene fluoride (VDF) and conduct high-throughput computational screening to identify top-performing MOFs with high VDF adsorption capacities. Quantitative structure-property relationships are analyzed via machine learning models on the combinations of geometric, chemical, and topological features, followed by feature importance analysis to probe the effects of these features on VDF adsorption. Finally, from detailed structural analysis via radial distribution functions and spatial densities, we elucidate the significance of different interaction modes between VDF and metal nodes in top-performing MOFs. By synergizing force-field development, computational screening, and machine learning, our findings provide microscopic insights into VDF adsorption in MOFs that will advance the development of new nanoporous materials for high-performance VDF storage or capture.

Molecular Interactions of Perfluorinated and Branched Fluorine-Free Surfactants at Interfaces: Insights from a New Reliable Force Field.

Per- and polyfluoroalkyl substances (PFAS) constitute a class of synthetic compounds with exceptional interfacial properties. Their widespread use in many industrial applications and consumer products, combined with their remarkable chemical and thermal stability, has led to their ubiquitous presence in environmental matrices, including surface water and groundwater. To replace PFAS with fluorine-free surfactants, it is necessary first to develop a deep molecular-level understanding of the mechanisms responsible for the exceptional properties of PFAS. For instance, it has been shown that fluorine-free surfactants with highly branched or methylated chains can achieve low surface tensions at air-water interfaces and can provide highly hydrophobic surface coatings. Although molecular simulations combined with experiments are promising for uncovering these mechanisms, the reliability of simulation results depends strongly on the accuracy of the force fields implemented. At the moment, atomistic force fields are not available to describe PFAS in a variety of environments. Ab initio methods could help fill this knowledge gap, but they are computationally demanding. As an alternative, ab initio calculations could be used to develop accurate force fields for atomistic simulations. In this work, a new algorithm is proposed, which, built from accurate ab initio calculations, yields force fields for perfluorinated sulfonic and perfluoroalkyl acids. The accuracy of the new force field was benchmarked against solvation free energy and interfacial tension data. The new force fields were then used to probe the interfacial behavior of the PFAS surfactants. The interfacial properties observed in our simulations were compared with those manifested by two branched fluorine-free surfactants. The good agreement achieved with experiments and ab initio calculations suggests that the proposed protocol could be implemented to study other perfluorinated substances and help in the design of fluorine-free surfactants for targeted applications.

Accurate Structure Prediction for Cyclic Peptides Containing Proline Residues with High-Temperature Molecular Dynamics.

Cyclic peptides (CPs) are emerging as promising drug candidates. Numerous natural CPs and their analogs are effective therapeutics against various diseases. Notably, many of them contain peptidyl cis-prolyl bonds. Due to the high rotational barrier of peptide bonds, conventional molecular dynamics simulations struggle to effectively sample the cis/trans-isomerization of peptide bonds. Previous studies have highlighted the high accuracy of the residue-specific force field (RSFF) and the high sampling efficiency of high-temperature molecular dynamics (high-T MD). Herein, we propose a protocol that combines high-T MD with RSFF2C and a recently developed reweighting method based on probability densities for accurate structure prediction of proline-containing CPs. Our method successfully predicted 19 out of 23 CPs with the backbone rmsd < 1.0 Å compared to X-ray structures. Furthermore, we performed high-T MD and density reweighting on the sunflower trypsin inhibitor (SFTI-1)/trypsin complex to demonstrate its applicability in studying CP-complexes containing cis-prolines. Our results show that the conformation of SFTI-1 in aqueous solution is consistent with its bound conformation, potentially facilitating its binding.

Wild-Type α-Synuclein Structure and Aggregation: A Comprehensive Coarse-Grained and All-Atom Molecular Dynamics Study.

α-Synuclein (α-syn) is a 140 amino acid intrinsically disordered protein (IDP) and the primary component of cytotoxic oligomers implicated in the etiology of Parkinson's disease (PD). While IDPs lack a stable three-dimensional structure, they sample a heterogeneous ensemble of conformations that can, in principle, be assessed through molecular dynamics simulations. However, describing the structure and aggregation of large IDPs is challenging due to force field (FF) accuracy and sampling limitations. To cope with the latter, coarse-grained (CG) FFs emerge as a potential alternative at the expense of atomic detail loss. Whereas CG models can accurately describe the structure of the monomer, less is known about aggregation. The latter is key for assessing aggregation pathways and designing aggregation inhibitor drugs. Herein, we investigate the structure and dynamics of α-syn using different resolution CG (Martini3 and Sirah2) and all-atom (Amber99sb and Charmm36m) FFs to gain insight into the differences and resemblances between these models. The dependence of the magnitude of protein-water interactions and the putative need for enhanced sampling (replica exchange) methods in CG simulations are analyzed to distinguish between force field accuracy and sampling limitations. The stability of the CG models of an α-syn fibril was also investigated. Additionally, α-syn aggregation was studied through umbrella sampling for the CG models and CG/all-atom models for an 11-mer peptide (NACore) from an amyloidogenic domain of α-syn. Our results show that despite the α-syn structures of Martini3 and Sirah2 with enhanced protein-water interactions being similar, major differences exist concerning aggregation. The Martini3 fibril is not stable, and the binding free energy of α-syn and NACore is positive, opposite to Sirah2. Sirah2 peptides in a zwitterionic form, in turn, display termini interactions that are too strong, resulting in end-to-end orientation. Sirah2, with enhanced protein-water interactions and neutral termini, provides, however, a peptide aggregation free energy profile similar to that found with all-atom models. Overall, we find that Sirah2 with enhanced protein-water interactions is suitable for studying protein-protein and protein-drug aggregation.

Simulation-Guided Conformational Space Exploration to Assess Reactive Conformations of a Ribozyme.

Self-splicing ribozymes are small ribonucleic acid (RNA) enzymes that catalyze their own cleavage through a transphosphoesterification reaction. While this process is involved in some specific steps of viral RNA replication and splicing, it is also of importance in the context of the (putative) first autocatalytic RNA-based systems that could have preceded the emergence of modern life. The uncatalyzed phosphoester bond formation is thermodynamically very unfavorable, and many experimental studies have focused on understanding the molecular features of catalysis in these ribozymes. However, chemical reaction paths are short-lived and not easily characterized by experimental approaches, so molecular simulation approaches appear as an ideal tool to unveil the molecular details of the reaction. Here, we focus on the model hairpin ribozyme. We show that identifying a relevant initial conformation for reactivity studies, which is frequently overlooked in mixed quantum-classical studies that predominantly concentrate on the chemical reaction itself, can be highly challenging. These challenges stem from limitations in both available experimental structures (which are chemically altered to prevent self-cleavage) and the accuracy of force fields, together with the necessity for comprehensive sampling. We show that molecular dynamics simulations, combined with extensive conformational phase space exploration with Hamiltonian replica-exchange simulations, enable us to characterize the relevant conformational basins of the minimal hairpin ribozyme in the ligated state prior to self-cleavage. We find that what is usually considered a canonical reactive conformation with active site geometries and hydrogen-bond patterns that are optimal for the addition-elimination reaction with general acid/general base catalysis is metastable and only marginally populated. The thermodynamically stable conformation appears to be consistent with the expectations of a mechanism that does not require the direct participation of ribozyme residues in the reaction. While these observations may suffer from forcefield inaccuracies, all investigated forcefields lead to the same conclusions upon proper sampling, contrasting with previous investigations on shorter timescales suggesting that at least one reparametrization of the Amber99 forcefield allowed to stabilize aligned active site conformations. Our study demonstrates that identifying the most pertinent reactant state conformation holds equal importance alongside the accurate determination of the thermodynamics and kinetics of the chemical steps of the reaction.

Fluorinated Protein-Ligand Complexes: A Computational Perspective.

Fluorine is an element renowned for its unique properties. Its powerful capability to modulate molecular properties makes it an attractive substituent for protein binding ligands; however, the rational design of fluorination can be challenging with effects on interactions and binding energies being difficult to predict. In this Perspective, we highlight how computational methods help us to understand the role of fluorine in protein-ligand binding with a focus on molecular simulation. We underline the importance of an accurate force field, present fluoride channels as a showcase for biomolecular interactions with fluorine, and discuss fluorine specific interactions like the ability to form hydrogen bonds and interactions with aryl groups. We put special emphasis on the disruption of water networks and entropic effects.

Detailed Reaction Kinetics for Hydrocarbon Fuels: The Development and Application of the ReaxFFCHO-S22 Force Field for C/H/O Systems with Enhanced Accuracy.

Efficient and accurate reactive force fields (e.g., ReaxFF) are pivotal for large-scale atomistic simulations to comprehend microscopic combustion processes. ReaxFF has been extensively utilized to describe chemical reactions in condensed phases, but most existing ReaxFF models rely on quantum mechanical (QM) data nearly two decades old, particularly in hydrocarbon systems, constraining their accuracy and applicability. Addressing this gap, we introduce a reparametrized ReaxFFCHO-S22 for C/H/O systems, tailored for studying the pyrolysis and combustion of hydrocarbon fuel. Our approach involves high-level QM benchmarks and large database construction for C/H/O systems, global ReaxFF parameter optimization, and molecular dynamics simulations of typical hydrocarbon fuels. Density functional theory (DFT) computations utilized the M06-2X functional at the meta-generalized gradient approximation (meta-GGA) level with a large basis set (6-311++G**). Our new ReaxFFCHO-S22 model exhibits a minimum 10% enhancement in accuracy compared to the previous ReaxFF models for a large variety of hydrocarbon molecules. This advanced ReaxFFCHO-S22 not only enables efficient large-scale studies on the microscopic chemical reactions of more complex hydrocarbon fuel but also can extend to biofuels, energetic materials, polymers, and other pertinent systems, thus serving as a valuable tool for studying chemical reaction dynamics of the large-scale hydrocarbon condensed phase at an atomistic level.

Exploring the Limits of the Generalized CHARMM and AMBER Force Fields through Predictions of Hydration Free Energy of Small Molecules.

Accurate force field parameters, potential energy functions, and receptor-ligand models are essential for modeling the solvation and binding of drug-like molecules to a receptor. A large and ever-growing chemical space of medicinally relevant scaffolds has also required these factors, especially force field parameters, to be highly transferable. Generalized force fields such as the CHARMM General Force Field (CGenFF) and the generalized AMBER force field (GAFF) have accomplished this feat along with other contemporaneous ones like OPLS. Here, we analyze the limits in the parametrization of drug-like small molecules by CGenFF and GAFF in terms of the various functional groups represented within them. Specifically, we link the presence of specific functional groups to the error in the absolute hydration free energy of over 600 small molecules, predicted by alchemical free energy methods implemented in the CHARMM program. Our investigation reveals that molecules with (i) a nitro group in CGenFF and GAFF are, respectively, over- or undersolubilized in aqueous medium, (ii) amine groups are undersolubilized more so in CGenFF than in GAFF, and (iii) carboxyl groups are more oversolubilized in GAFF than in CGenFF. We present our analyses of the potential factors underlying these trends. We also showcase the use of a machine-learning-based approach combined with the SHapley Additive exPlanations framework to attribute these trends to specific functional groups, which can be easily adopted to explore the limits of other general force fields.

Complexity of many-body interactions in transition metals via machine-learned force fields from the TM23 data set

This work examines challenges associated with the accuracy of machine-learned force fields (MLFFs) for bulk solid and liquid phases of d-block elements. In exhaustive detail, we contrast the performance of force, energy, and stress predictions across the transition metals for two leading MLFF models: a kernel-based atomic cluster expansion method implemented using sparse Gaussian processes (FLARE), and an equivariant message-passing neural network (NequIP). Early transition metals present higher relative errors and are more difficult to learn relative to late platinum- and coinage-group elements, and this trend persists across model architectures. Trends in complexity of interatomic interactions for different metals are revealed via comparison of the performance of representations with different many-body order and angular resolution. Using arguments based on perturbation theory on the occupied and unoccupied d states near the Fermi level, we determine that the large, sharp d density of states both above and below the Fermi level in early transition metals leads to a more complex, harder-to-learn potential energy surface for these metals. Increasing the fictitious electronic temperature (smearing) modifies the angular sensitivity of forces and makes the early transition metal forces easier to learn. This work illustrates challenges in capturing intricate properties of metallic bonding with current leading MLFFs and provides a reference data set for transition metals, aimed at benchmarking the accuracy and improving the development of emerging machine-learned approximations.

Application of the CL&P-Pol polarizable force field in the determination of the excess chemical potential of thiophene within the [C4mim] [BF4, Cl, Br, CH3COO] ionic liquids

The combustion of fossil fuels is an important source of air pollution due to the presence of sulfur-based compounds. In this regard, a reliable atomistic forcefield is required in order to provide insights on solute-solvent interactions, and in designing high aromatic sulfur extracting solvents. In the present work, the excess chemical potentials of thiophene within four 1-butyl-3-methylimidazolium-based ionic liquids were evaluated through the free energy perturbation with replica exchange molecular dynamics methodology by using the CL&Pol polarizable force fields based on Drude oscillators. First, in order to validate the accuracy of the polarizable force fields, a series of thermodynamic properties of pure ionic liquids such as diffusion coefficient, liquid density, dielectric constant, and heat of vaporization were computed using the CL&Pol force field and compared against experimental values. The results from the calculated thiophene excess chemical potential values, indicated that the [C4mim][CH3COO] ionic liquid presents the more favorable excess chemical potential (higher solubility) with respect to the thiophene molecule at both 300 K and 343 K in comparison to the other ionic liquids studied. The structural analysis revealed that the ionic liquid anion interacts more closely and with well-defined RDF peaks with the thiophene molecule, the larger anions present higher surrounding particle densities with the thiophene molecules (solvation shell), while the monoatomic anions interact preferentially with the hydrogens close to the sulfur thiophene atom; In contrast no marked solvation layers were observed with the ionic liquid cations.

Molecular dynamics simulation of methane hydrates: Prediction of the phase equilibria using extracted microscopic parameters from SAFT-VR Mie EOS

This paper explores the three-phase equilibrium of structure-I methane hydrate by utilizing the extracted Mie potential parameters for non-bonded molecular interactions in the LAMMPS molecular dynamics simulator. An in-house code was developed to numerically solve the SAFT-VR Mie equation of state (EOS). To reduce numerical errors, unlike common practices, we incorporated all partial derivatives analytically in the code. The microscopic simulation parameters for methane and water molecules were selected so that saturation pressure and liquid density data from EOS were fitted on experimental data. Using this approach, the three-phase (V−Lw−H) equilibrium temperatures for pressure values of 30, 70, 100, and 300 bar were calculated. The solubility of methane in the aqueous phase at three-phase equilibrium was also determined. The results were validated against available laboratory data. To evaluate the performance of Mie potential parameters, the MD simulations were repeated using three well-known force fields for methane molecules, namely, Optimized Potentials for Liquid Simulations-United Atom (OPLS-UA), Universal force field (UFF), and General Amber force field (GAFF). The TIP4P/Ice force field was used for water molecules in all cases. The comparison with experimental data reveals that while the accuracy of the UFF force field is the highest, our simulation approach outperforms OPLS-UA and GAFF force fields in predicting methane solubility in the aqueous phase. Besides molecular dynamics simulation, the van der Waals-Platteeuw model was combined with the SAFT VR Mie EOS to account for the hydrate phase. In our case study, the SAFT-VR Mie + vdW-P model proved to be more accurate in predicting the three-phase equilibrium data compared to the MD approach.

Balancing Group I Monatomic Ion-Polar Compound Interactions for Condensed Phase Simulation in the Polarizable Drude Force Field.

Molecular dynamics (MD) simulations are a commonly used method for investigating molecular behavior at the atomic level. Achieving reliable MD simulation results necessitates the use of an accurate force field. In the present work, we present a protocol to enhance the quality of group 1 monatomic ions (specifically Li+, Na+, K+, Rb+, and Cs+) with respect to their interactions with common polar model compounds in biomolecules in condensed phases in the context of the Drude polarizable force field. Instead of adjusting preexisting individual parameters for ions, model compounds, and water, we employ atom-pair specific Lennard-Jones (LJ) (known as NBFIX in CHARMM) and through-space Thole dipole screening (NBTHOLE) terms to fine-tune the balance of ion-model compound, ion-water, and model compound-water interactions. This involved establishing a protocol for the optimization of NBFIX and NBTHOLE parameters targeting the difference between molecular mechanical (MM) and quantum mechanical (QM) potential energy scans (PES). It is shown that targeting PES involving complexes that include multiple model compounds and/or ions as trimers and tetramers yields parameters that produce condensed phase properties in agreement with experimental data. Validation of this protocol involved the reproduction of experimental thermodynamic benchmarks, including solvation free energies of ions in methanol and N-methylacetamide, osmotic pressures, ionic conductivities, and diffusion coefficients within the condensed phase. These results show the importance of including more complex ion-model compound complexes beyond dimers in the QM target data to account for many-body effects during parameter fitting. The presented parameters represent a significant refinement of the Drude polarizable force field, which will lead to improved accuracy for modeling ion-biomolecular interactions.

Accurate machine learning force fields via experimental and simulation data fusion

Machine Learning (ML)-based force fields are attracting ever-increasing interest due to their capacity to span spatiotemporal scales of classical interatomic potentials at quantum-level accuracy. They can be trained based on high-fidelity simulations or experiments, the former being the common case. However, both approaches are impaired by scarce and erroneous data resulting in models that either do not agree with well-known experimental observations or are under-constrained and only reproduce some properties. Here we leverage both Density Functional Theory (DFT) calculations and experimentally measured mechanical properties and lattice parameters to train an ML potential of titanium. We demonstrate that the fused data learning strategy can concurrently satisfy all target objectives, thus resulting in a molecular model of higher accuracy compared to the models trained with a single data source. The inaccuracies of DFT functionals at target experimental properties were corrected, while the investigated off-target properties were affected only mildly and mostly positively. Our approach is applicable to any material and can serve as a general strategy to obtain highly accurate ML potentials.

McGFN-FF: an accurate force field for optimization and energetic screening of molecular crystals

Abstract This work presents a periodic extension of the GFN-FF force field for molecular crystals named mcGFN-FF. Non-covalent interactions in the force field are adjusted to reduce the systematic overbinding of the original, molecular version for molecular crystals. A diverse set of molecular crystal benchmarks for lattice energies and unit cell volumes is studied. The modified force field shows good results with a mean absolute relative deviation (MARD) of 19.9 % for lattice energies and 10.0 % for unit cell volumes. In many cases, mcGFN-FF approaches the accuracy of the GFN1-xTB quantum chemistry method which has an MARD of 18.7 % for lattice energies and 6.2 % for unit cell volumes. Further, the newly compiled mcVOL22 benchmark set is presented which features r2SCAN-D4/900 eV DFT reference volumes for molecular crystals with phosphorus-, sulfur-, and chlorine-containing compounds of various sizes. Overall, the mcGFN-FF poses an efficient tool for the optimization and energetic screening of molecular crystals containing elements up to radon.

Ion Transport in Polymer Electrolytes: Building New Bridges between Experiment and Molecular Simulation.

ConspectusPolymer electrolytes constitute a promising type of material for solid-state batteries. However, one of the bottlenecks for their practical implementation lies in the transport properties, often including restricted Li+ self-diffusion and conductivity and low cationic transference numbers. This calls for a molecular understanding of ion transport in polymer electrolytes in which molecular dynamics (MD) simulation can provide both new physical insights and quantitative predictions. Although efforts have been made in this area and qualitative pictures have emerged, direct and quantitative comparisons between experiment and simulation remain challenging because of the lack of a unified theoretical framework to connect them.In our work, we show that by computing the glass transition temperature (Tg) of the model system and using the normalized inverse temperature 1000/(T - Tg + 50), the Li+ self-diffusion coefficient can be compared quantitatively between MD simulations and experiments. This allows us to disentangle the effects of Tg and the polymer dielectric environment on ion conduction in polymer electrolytes, giving rise to the identification of an optimal solvating environment for fast ion conduction.Unlike Li+ self-diffusion coefficients and ionic conductivity, the transference number, which describes the fraction of current carried by Li+ ions, depends on the boundary conditions or the reference frame (RF). This creates a non-negligible gap when comparing experiment and simulation because the fluxes in the experimental measurements and in the linear response theory used in MD simulation are defined in different RFs. We show that by employing the Onsager theory of ion transport and applying a proper RF transformation, a much better agreement between experiment and simulation can be achieved for the PEO-LiTFSI system. This further allows us to derive the theoretical expression for the Bruce-Vincent transference number in terms of the Onsager coefficients and make a direct comparison to experiments. Since the Bruce-Vincent method is widely used to extract transference numbers from experimental data, this opens the door to calibrating MD simulations via reproducing the Bruce-Vincent transference number and using MD simulations to predict the true transference number.In addition, we also address several open questions here such as the time-scale effects on the ion-pairing phenomenon, the consistency check between different types of experiments, the need for more accurate force fields used in MD simulations, and the extension to multicomponent systems. Overall, this Account focuses on building new bridges between experiment and simulation for quantitative comparison, warnings of pitfalls when comparing apples and oranges, and clarifying misconceptions. From a physical chemistry point of view, it connects to concentrated solution theory and provides a unified theoretical framework that can maximize the power of MD simulations. Therefore, this Account will be useful for the electrochemical energy storage community at large and set examples of how to approach experiments from theory and simulation (and vice versa).

Thermal transports of 2D phosphorous carbides by machine learning molecular dynamics simulations

Carbon phosphide is a newly discovered two-dimensional semiconductor material which wrinkles and has a significant carrier mobility. Due to lack an accurate force field, the use of molecular dynamics to study its phonon-dominated thermal conductivity which lead to inaccurate results. At present, the use of machine learning to construct a high-precision force field has become the mainstream research method to solve this problem. The main work of this study is to construct a comprehensive training sets for Phosphorus-Doped Graphene (PCn) (n = 3, 5, 6) and to use the fitted potential to calculate the related thermal properties. The research found that (PC5) exhibited anisotropic behavior, with a thermal conductivity of 106.6 Wm−1K−1 in the y-direction and 63.6 Wm−1K−1 in the x-direction. In comparison, (PC6) and (PC3) showed isotropic behavior, with thermal conductivity of approximately 104 Wm−1K−1 and 76.83 Wm−1K−1, respectively. Compared to monolayer graphene, the lower thermal conductivity of PCn is mainly attributed to phonon-phonon scattering effects, which are limited by the regular wrinkled structure. Additionally, low-frequency phonon have been found to have a significant impact on the thermal performance of PCn. Furthermore, we investigated the influence of uniaxial strain on the PC6 and observed an increase in the thermal conductivity with increasing strain. This study used key computational and analytical techniques, including phonon dispersion relations, homogeneous nonequilibrium molecular dynamics method, spectral thermal conductivity analysis. These findings provide a theoretical basis for understanding the thermal transport properties of PCn and will guide its potential applications value.

Reparameterization of Polarizable Force Fields for Studying Ion Transfer across Liquid-Liquid Interfaces.

We have developed a general scheme for refining classical polarizable molecular dynamics (MD) force fields that can accurately describe the molecular interactions in systems with liquid-liquid interfaces. While ab initio MD (AIMD) simulations can naturally describe molecular interactions, they are often so computationally expensive that simulating large system sizes and/or long time scales is usually infeasible. To resolve this, we parameterized efficient and accurate classical polarizable force fields that use AIMD reference data by minimizing both the relative entropy and the root mean squared deviation in atomic forces. We utilized our new multiscale models to study chloride ion transfer across the water-dichloromethane (DCM) interface with and without the tetraethylammonium cation as the phase-transfer catalyst. Our calculated free-energy barrier for the water-DCM interface is consistent with the other reported simulation results. We further analyzed the ion-transfer process by studying the hydration shell structures around the chloride ion and the ion-pair formation to better understand the mechanism. We observed that electronic polarizability is an important factor for the studied phase-transfer catalyst to lower the free-energy barrier of the ion transfer. Using the water-benzene interface system as an additional example, we show that our parameterization scheme provides a general route for modeling different liquid-liquid interface systems even when the experimental data or force field parameters are not readily available.

Machine Learning Methods as a Cost-Effective Alternative to Physics-Based Binding Free Energy Calculations.

The rank ordering of ligands remains one of the most attractive challenges in drug discovery. While physics-based in silico binding affinity methods dominate the field, they still have problems, which largely revolve around forcefield accuracy and sampling. Recent advances in machine learning have gained traction for protein-ligand binding affinity predictions in early drug discovery programs. In this article, we perform retrospective binding free energy evaluations for 172 compounds from our internal collection spread over four different protein targets and five congeneric ligand series. We compared multiple state-of-the-art free energy methods ranging from physics-based methods with different levels of complexity and conformational sampling to state-of-the-art machine-learning-based methods that were available to us. Overall, we found that physics-based methods behaved particularly well when the ligand perturbations were made in the solvation region, and they did not perform as well when accounting for large conformational changes in protein active sites. On the other end, machine-learning-based methods offer a good cost-effective alternative for binding free energy calculations, but the accuracy of their predictions is highly dependent on the experimental data available for training the model.