Enhancing Protein-Ligand Binding Affinity Predictions Using Neural Network Potentials.

This letter gives results on improving protein-ligand binding affinity predictions based on molecular dynamics simulations using machine learning potentials with a hybrid neural network potential and molecular mechanics methodology (NNP/MM). We compute relative binding free energies with the Alchemical Transfer Method and validate its performance against established benchmarks and find significant enhancements compared with conventional MM force fields like GAFF2.

NNP/MM: Accelerating Molecular Dynamics Simulations with Machine Learning Potentials and Molecular Mechanics.

Machine learning potentials have emerged as a means to enhance the accuracy of biomolecular simulations. However, their application is constrained by the significant computational cost arising from the vast number of parameters compared with traditional molecular mechanics. To tackle this issue, we introduce an optimized implementation of the hybrid method (NNP/MM), which combines a neural network potential (NNP) and molecular mechanics (MM). This approach models a portion of the system, such as a small molecule, using NNP while employing MM for the remaining system to boost efficiency. By conducting molecular dynamics (MD) simulations on various protein-ligand complexes and metadynamics (MTD) simulations on a ligand, we showcase the capabilities of our implementation of NNP/MM. It has enabled us to increase the simulation speed by ∼5 times and achieve a combined sampling of 1 µs for each complex, marking the longest simulations ever reported for this class of simulations.

Psi4 1.4: Open-source software for high-throughput quantum chemistry.

PSI4 is a free and open-source ab initio electronic structure program providing implementations of Hartree-Fock, density functional theory, many-body perturbation theory, configuration interaction, density cumulant theory, symmetry-adapted perturbation theory, and coupled-cluster theory. Most of the methods are quite efficient, thanks to density fitting and multi-core parallelism. The program is a hybrid of C++ and Python, and calculations may be run with very simple text files or using the Python API, facilitating post-processing and complex workflows; method developers also have access to most of PSI4's core functionalities via Python. Job specification may be passed using The Molecular Sciences Software Institute (MolSSI) QCSCHEMA data format, facilitating interoperability. A rewrite of our top-level computation driver, and concomitant adoption of the MolSSI QCARCHIVE INFRASTRUCTURE project, makes the latest version of PSI4 well suited to distributed computation of large numbers of independent tasks. The project has fostered the development of independent software components that may be reused in other quantum chemistry programs.

A Scalable Molecular Force Field Parameterization Method Based on Density Functional Theory and Quantum-Level Machine Learning.

Fast and accurate molecular force field (FF) parameterization is still an unsolved problem. Accurate FF are not generally available for all molecules, like novel druglike molecules. While methods based on quantum mechanics (QM) exist to parameterize them with better accuracy, they are computationally expensive and slow, which limits applicability to a small number of molecules. Here, we present an automated FF parameterization method which can utilize either density functional theory (DFT) calculations or approximate QM energies produced by different neural network potentials (NNPs), to obtain improved parameters for molecules. We demonstrate that for the case of torchani-ANI-1x NNP, we can parameterize small molecules in a fraction of time compared with an equivalent parameterization using DFT QM calculations while producing more accurate parameters than FF (GAFF2). We expect our method to be of critical importance in computational structure-based drug discovery (SBDD). The current version is available at PlayMolecule ( www.playmolecule.org ) and implemented in HTMD, allowing to parameterize molecules with different QM and NNP options.

Replica state exchange metadynamics for improving the convergence of free energy estimates.

Metadynamics (MTD) is a powerful enhanced sampling method for systems with rugged energy landscapes. It constructs a bias potential in a predefined collective variable (CV) space to overcome barriers between metastable states. In bias-exchange MTD (BE-MTD), multiple replicas approximate the CV space by exchanging bias potentials (replica conditions) with the Metropolis-Hastings (MH) algorithm. We demonstrate that the replica-exchange rates and the convergence of free energy estimates of BE-MTD are improved by introducing the infinite swapping (IS) or the Suwa-Todo (ST) algorithms. Conceptually, IS and ST perform transitions in a replica state space rather than exchanges in a replica condition space. To emphasize this, the proposed scheme is called the replica state exchange MTD (RSE-MTD). Benchmarks were performed with alanine polypeptides in vacuum and water. For the systems tested in this work, there is no significant performance difference between IS and ST.

Enhancing Protein-Ligand Binding Affinity Predictions using Neural Network Potentials.

This letter gives results on improving protein-ligand binding affinity predictions based on molecular dynamics simulations using machine learning potentials with a hybrid neural network potential and molecular mechanics methodology (NNP/MM). We compute relative binding free energies (RBFE) with the Alchemical Transfer Method (ATM) and validate its performance against established benchmarks and find significant enhancements compared to conventional MM force fields like GAFF2.

TorchMD-Net 2.0: Fast Neural Network Potentials for Molecular Simulations.

Achieving a balance between computational speed, prediction accuracy, and universal applicability in molecular simulations has been a persistent challenge. This paper presents substantial advancements in TorchMD-Net software, a pivotal step forward in the shift from conventional force fields to neural network-based potentials. The evolution of TorchMD-Net into a more comprehensive and versatile framework is highlighted, incorporating cutting-edge architectures such as TensorNet. This transformation is achieved through a modular design approach, encouraging customized applications within the scientific community. The most notable enhancement is a significant improvement in computational efficiency, achieving a very remarkable acceleration in the computation of energy and forces for TensorNet models, with performance gains ranging from 2× to 10× over previous, nonoptimized, iterations. Other enhancements include highly optimized neighbor search algorithms that support periodic boundary conditions and smooth integration with existing molecular dynamics frameworks. Additionally, the updated version introduces the capability to integrate physical priors, further enriching its application spectrum and utility in research. The software is available at https://github.com/torchmd/torchmd-net.

TorchMD-Net 2.0: Fast Neural Network Potentials for Molecular Simulations.

Achieving a balance between computational speed, prediction accuracy, and universal applicability in molecular simulations has been a persistent challenge. This paper presents substantial advancements in the TorchMD-Net software, a pivotal step forward in the shift from conventional force fields to neural network-based potentials. The evolution of TorchMD-Net into a more comprehensive and versatile framework is highlighted, incorporating cutting-edge architectures such as TensorNet. This transformation is achieved through a modular design approach, encouraging customized applications within the scientific community. The most notable enhancement is a significant improvement in computational efficiency, achieving a very remarkable acceleration in the computation of energy and forces for Tensor-Net models, with performance gains ranging from 2x to 10x over previous, non-optimized, iterations. Other enhancements include highly optimized neighbor search algorithms that support periodic boundary conditions and smooth integration with existing molecular dynamics frameworks. Additionally, the updated version introduces the capability to integrate physical priors, further enriching its application spectrum and utility in research. The software is available at https://github.com/torchmd/torchmd-net.

OpenMM 8: Molecular Dynamics Simulation with Machine Learning Potentials.

Machine learning plays an important and growing role in molecular simulation. The newest version of the OpenMM molecular dynamics toolkit introduces new features to support the use of machine learning potentials. Arbitrary PyTorch models can be added to a simulation and used to compute forces and energy. A higher-level interface allows users to easily model their molecules of interest with general purpose, pretrained potential functions. A collection of optimized CUDA kernels and custom PyTorch operations greatly improves the speed of simulations. We demonstrate these features in simulations of cyclin-dependent kinase 8 (CDK8) and the green fluorescent protein chromophore in water. Taken together, these features make it practical to use machine learning to improve the accuracy of simulations with only a modest increase in cost.

SPICE, A Dataset of Drug-like Molecules and Peptides for Training Machine Learning Potentials.

Machine learning potentials are an important tool for molecular simulation, but their development is held back by a shortage of high quality datasets to train them on. We describe the SPICE dataset, a new quantum chemistry dataset for training potentials relevant to simulating drug-like small molecules interacting with proteins. It contains over 1.1 million conformations for a diverse set of small molecules, dimers, dipeptides, and solvated amino acids. It includes 15 elements, charged and uncharged molecules, and a wide range of covalent and non-covalent interactions. It provides both forces and energies calculated at the ωB97M-D3(BJ)/def2-TZVPPD level of theory, along with other useful quantities such as multipole moments and bond orders. We train a set of machine learning potentials on it and demonstrate that they can achieve chemical accuracy across a broad region of chemical space. It can serve as a valuable resource for the creation of transferable, ready to use potential functions for use in molecular simulations.

OpenMM 8: Molecular Dynamics Simulation with Machine Learning Potentials.

Machine learning plays an important and growing role in molecular simulation. The newest version of the OpenMM molecular dynamics toolkit introduces new features to support the use of machine learning potentials. Arbitrary PyTorch models can be added to a simulation and used to compute forces and energy. A higher-level interface allows users to easily model their molecules of interest with general purpose, pretrained potential functions. A collection of optimized CUDA kernels and custom PyTorch operations greatly improves the speed of simulations. We demonstrate these features on simulations of cyclin-dependent kinase 8 (CDK8) and the green fluorescent protein (GFP) chromophore in water. Taken together, these features make it practical to use machine learning to improve the accuracy of simulations at only a modest increase in cost.

Exceptionally low shear modulus in a prototypical imidazole-based metal-organic framework.

Using Brillouin scattering, we measured the single-crystal elastic constants (C(ij)'s) of a prototypical metal-organic framework (MOF): zeolitic imidazolate framework (ZIF)-8 [Zn(2-methylimidazolate)(2)], which adopts a zeolitic sodalite topology and exhibits large porosity. Its C(ij)'s under ambient conditions are (in GPa) C(11)=9.522(7), C(12)=6.865(14), and C(44)=0.967(4). Tensorial analysis of the C(ij)'s reveals the complete picture of the anisotropic elasticity in cubic ZIF-8. We show that ZIF-8 has a remarkably low shear modulus G(min) < or approximately 1 GPa, which is the lowest yet reported for a single-crystalline extended solid. Using ab initio calculations, we demonstrate that ZIF-8's C(ij)'s can be reliably predicted, and its elastic deformation mechanism is linked to the pliant ZnN(4) tetrahedra. Our results shed new light on the role of elastic constants in establishing the structural stability of MOF materials and thus their suitability for practical applications.

Mechanical properties of dense zeolitic imidazolate frameworks (ZIFs): a high-pressure X-ray diffraction, nanoindentation and computational study of the zinc framework Zn(Im)2, and its lithium-boron analogue, LiB(Im)4.

The dense, anhydrous zeolitic imidazolate frameworks (ZIFs), Zn(Im)(2) (1) and LiB(Im)(4) (2), adopt the same zni topology and differ only in terms of the inorganic species present in their structures. Their mechanical properties (specifically the Young's and bulk moduli, along with the hardness) have been elucidated by using high pressure, synchrotron X-ray diffraction, density functional calculations and nanoindentation studies. Under hydrostatic pressure, framework 2 undergoes a phase transition at 1.69 GPa, which is somewhat higher than the transition previously reported in 1. The Young's modulus (E) and hardness (H) of 1 (E≈8.5, H≈1 GPa) is substantially higher than that of 2 (E≈3, H≈0.1 GPa), whilst its bulk modulus is relatively lower (≈14 GPa cf. ≈16.6 GPa). The heavier, zinc-containing material was also found to be significantly harder than its light analogue. The differential behaviour of the two materials is discussed in terms of the smaller pore volume of 2 and the greater flexibility of the LiN(4) tetrathedron compared with the ZnN(4) and BN(4) units.

Neural Network and Nearest Neighbor Algorithms for Enhancing Sampling of Molecular Dynamics.

The free energy calculations of complex chemical and biological systems with molecular dynamics (MD) are inefficient due to multiple local minima separated by high-energy barriers. The minima can be escaped using an enhanced sampling method such as metadynamics, which apply bias (i.e., importance sampling) along a set of collective variables (CV), but the maximum number of CVs (or dimensions) is severely limited. We propose a high-dimensional bias potential method (NN2B) based on two machine learning algorithms: the nearest neighbor density estimator (NNDE) and the artificial neural network (ANN) for the bias potential approximation. The bias potential is constructed iteratively from short biased MD simulations accounting for correlation among CVs. Our method is capable of achieving ergodic sampling and calculating free energy of polypeptides with up to 8-dimensional bias potential.

Enhanced Conformational Sampling of N-Glycans in Solution with Replica State Exchange Metadynamics.

Molecular dynamics (MD) simulation of a N-glycan in solution is challenging because of high-energy barriers of the glycosidic linkages, functional group rotational barriers, and numerous intra- and intermolecular hydrogen bonds. In this study, we apply different enhanced conformational sampling approaches, namely, metadynamics (MTD), the replica-exchange MD (REMD), and the recently proposed replica state exchange MTD (RSE-MTD), to a N-glycan in solution and compare the conformational sampling efficiencies of the approaches. MTD helps to cross the high-energy barrier along the ω angle by utilizing a bias potential, but it cannot enhance sampling of the other degrees of freedom. REMD ensures moderate-energy barrier crossings by exchanging temperatures between replicas, while it hardly crosses the barriers along ω. In contrast, RSE-MTD succeeds to cross the high-energy barrier along ω as well as to enhance sampling of the other degrees of freedom. We tested two RSE-MTD schemes: in one scheme, 64 replicas were simulated with the bias potential along ω at different temperatures, while simulations of four replicas were performed with the bias potentials for different CVs at 300 K. In both schemes, one unbiased replica at 300 K was included to compute conformational properties of the glycan. The conformational sampling of the former is better than the other enhanced sampling methods, while the latter shows reasonable performance without spending large computational resources. The latter scheme is likely to be useful when a N-glycan-attached protein is simulated.

Coarse graining of force fields for metal-organic frameworks.

We have adapted our genetic algorithm based optimization approach, originally developed to generate force field parameters from quantum mechanic reference data, to derive a first coarse grained force field for a MOF, taking the atomistic MOF-FF as a reference. On the example of the copper paddle-wheel based HKUST-1, a maximally coarse grained model, using a single bead for each three and four coordinated vertex, was developed as a proof of concept. By adding non-bonded interactions with a modified Buckingham potential, the resulting MOF-FF-CGNB is able to predict local deformation energies of the building blocks as well as bulk properties like the tbovs.pto energy difference or elastic constants in a semi-quantitative way. As expected, the negative thermal expansion of HKUST-1 is not reproduced by the maximally coarse grained model. At the expense of atomic resolution, substantially larger systems (up to tens of nanometers in size) can be simulated with respect to structural and mechanical properties, bridging the gap to the mesoscale. As an example the deformation of the [111] surface of HKUST-1 by a "tip" could be computed without artifacts from periodic images.

Flexibility and swing effect on the adsorption of energy-related gases on ZIF-8: combined experimental and simulation study.

ZIF-8, a prototypical zeolitic porous coordination polymer, prepared via the self-assembly of tetrahedral atoms (e.g. Zn and Co) and organic imidazolate linkers, presents large cavities which are interconnected by narrow windows that allow, in principle, molecular sieving. However, ZIF-8 shows flexibility due to the swing of the imidazolate linkers, which results in the adsorption of molecules which are too large to fit through the narrow window. In this work, we assess the impact of this flexibility, previously only observed for nitrogen, and the level of agreement between the experimental and simulated isotherms of different energy-related gases on ZIF-8 (CO(2), CH(4) and alkanes). We combine experimental gas adsorption with GCMC simulations, using generic and adjusted force fields and DFT calculations with the Grimme dispersion correction. By solely adapting the UFF force field to reduce the Lennard-Jones parameter ε, we achieve excellent agreement between the simulated and experimental results not only for ZIF-8 but also for ZIF-20, where the transferability of the adapted force field is successfully tested. Regarding ZIF-8, we show that two different structural configurations are needed to properly describe the adsorption performance of this material, demonstrating that ZIF-8 is undergoing a structural change during gas adsorption. DFT calculations with the Grimme dispersion correction are consistent with the GCMC and experimental observations, illustrating the thermodynamics of the CH(4) adsorption sites and confirming the existence of a new adsorption site with a high binding energy within the 4-ring window of ZIF-8.