DeePMD-GNN: A DeePMD-kit Plugin for External Graph Neural Network Potentials DOI
Jinzhe Zeng, Timothy J. Giese, Duo Zhang

et al.

Journal of Chemical Information and Modeling, Journal Year: 2025, Volume and Issue: unknown

Published: March 27, 2025

Machine learning potentials (MLPs) have revolutionized molecular simulation by providing efficient and accurate models for predicting atomic interactions. MLPs continue to advance had profound impact in applications that include drug discovery, enzyme catalysis, materials design. The current landscape of MLP software presents challenges due the limited interoperability between packages, which can lead inconsistent benchmarking practices necessitates separate interfaces with dynamics (MD) software. To address these issues, we present DeePMD-GNN, a plugin DeePMD-kit framework extends its capabilities support external graph neural network (GNN) potentials.DeePMD-GNN enables seamless integration popular GNN-based models, such as NequIP MACE, within ecosystem. Furthermore, new infrastructure allows GNN be used combined quantum mechanical/molecular mechanical (QM/MM) using range corrected ΔMLP formalism.We demonstrate application DeePMD-GNN performing benchmark calculations NequIP, DPA-2 developed under consistent training conditions ensure fair comparison.

Language: Английский

Leveraging Machine Learning Potentials for In-Situ Searching of Active sites in Heterogeneous Catalysis DOI Creative Commons

Xiran Cheng,

Chenyu Wu,

Jiayan Xu

et al.

Precision Chemistry, Journal Year: 2024, Volume and Issue: 2(11), P. 570 - 586

Published: Sept. 11, 2024

This Perspective explores the integration of machine learning potentials (MLPs) in research heterogeneous catalysis, focusing on their role identifying

Language: Английский

Citations

6
Computational Tools for Handling Molecular Clusters: Configurational Sampling, Storage, Analysis, and Machine Learning DOI Creative Commons
Jakub Kubečka, Vitus Besel, Ivo Neefjes

et al.

ACS Omega, Journal Year: 2023, Volume and Issue: 8(47), P. 45115 - 45128

Published: Nov. 14, 2023

Computational modeling of atmospheric molecular clusters requires a comprehensive understanding their complex configurational spaces, interaction patterns, stabilities against fragmentation, and even dynamic behaviors. To address these needs, we introduce the Jammy Key framework, collection automated scripts that facilitate streamline cluster workflows. handles file manipulations between varieties integrated third-party programs. The framework is divided into three main functionalities: (1) for sampling (JKCS) to perform systematic clusters, (2) quantum chemistry (JKQC) analyze commonly used output files database construction, handling, analysis, (3) machine learning (JKML) manage methods in optimizing modeling. This automation utilization significantly reduces manual labor, greatly speeds up search configurations, thus increases number systems can be studied. Following example Atmospheric Cluster Database (ACDB) Elm (ACS Omega, 4, 10965–10984, 2019), modeled our group using have been stored an improved online GitHub repository named ACDB 2.0. In this work, present package alongside its assorted applications, which underline versatility. Using several illustrative examples, discuss how choose appropriate combinations methodologies treating particular types, including reactive, multicomponent, charged, or radical as well containing flexible multiconformer monomers heavy atoms. Finally, detailed tools acid–base clusters.

Language: Английский

Citations

11
Accelerating Molecular Vibrational Spectra Simulations with a Physically Informed Deep Learning Model DOI
Yuzhuo Chen, Sebastian V. Pios, Maxim F. Gelin

et al.

Journal of Chemical Theory and Computation, Journal Year: 2024, Volume and Issue: 20(11), P. 4703 - 4710

Published: June 3, 2024

In recent years, machine learning (ML) surrogate models have emerged as an indispensable tool to accelerate simulations of physical and chemical processes. However, there is still a lack ML that can accurately predict molecular vibrational spectra. Here, we present highly efficient multitask model termed Vibrational Spectra Neural Network (VSpecNN), calculate infrared (IR) Raman spectra based on dipole moments polarizabilities obtained on-the-fly via ML-enhanced dynamics simulations. The methodology applied pyrazine, prototypical polyatomic chromophore. VSpecNN-predicted energies are well within the accuracy (1 kcal/mol), errors for forces only half those from popular high-performance model. Compared ab initio reference, frequencies IR differ by less than 5.87 cm–1, intensities depolarization ratios reproduced. VSpecNN developed in this work highlights importance constructing accurate neural network potentials predicting

Language: Английский

Citations

4
FeNNol: An efficient and flexible library for building force-field-enhanced neural network potentials DOI
Thomas Plé,

Olivier Adjoua,

Louis Lagardère

et al.

The Journal of Chemical Physics, Journal Year: 2024, Volume and Issue: 161(4)

Published: July 25, 2024

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

Language: Английский

Citations

4
SpaiNN: equivariant message passing for excited-state nonadiabatic molecular dynamics DOI Creative Commons
Sascha Mausenberger, Carolin Müller, Alexandre Tkatchenko

et al.

Chemical Science, Journal Year: 2024, Volume and Issue: 15(38), P. 15880 - 15890

Published: Jan. 1, 2024

S pai NN employs ch N et P ack to train electronic properties across various potential energy curves, including energies, gradients, and couplings, while integrating with SHARC for excited state molecular dynamics simulations.

Language: Английский

Citations

4
Generative diffusion model for surface structure discovery DOI

Nikolaj Rønne,

Alán Aspuru‐Guzik, Bjørk Hammer

et al.

Physical review. B./Physical review. B, Journal Year: 2024, Volume and Issue: 110(23)

Published: Dec. 24, 2024

Language: Английский

Citations

4
EOSnet: Embedded Overlap Structures for Graph Neural Networks in Predicting Material Properties DOI
Shuo Tao, Li Zhu

The Journal of Physical Chemistry Letters, Journal Year: 2025, Volume and Issue: 16(3), P. 717 - 724

Published: Jan. 11, 2025

Graph Neural Networks (GNNs) have emerged as powerful tools for predicting material properties, yet they often struggle to capture many-body interactions and require extensive manual feature engineering. Here, we present EOSnet (Embedded Overlap Structures Networks), a novel approach that addresses these limitations by incorporating Gaussian Matrix (GOM) fingerprints node features within the GNN architecture. Unlike models rely on explicit angular terms or human-engineered features, efficiently encodes through orbital overlap matrices, providing rotationally invariant transferable representation of atomic environments. The model demonstrates superior performance across various prediction tasks materials' achieving particularly notable results in properties sensitive interactions. For band gap prediction, achieves mean absolute error 0.163 eV, surpassing previous state-of-the-art models. also excels mechanical classifying materials, with 97.7% accuracy metal/nonmetal classification. These demonstrate embedding GOM into enhances ability GNNs complex interactions, making tool discovery property prediction.

Language: Английский

Citations

0
chemtrain: Learning Deep Potential Models via Automatic Differentiation and Statistical Physics DOI Creative Commons

Paul Fuchs,

Stephan Thaler, Sebastien Röcken

et al.

Computer Physics Communications, Journal Year: 2025, Volume and Issue: unknown, P. 109512 - 109512

Published: Jan. 1, 2025

Language: Английский

Citations

0
Neural Network Potential with Multiresolution Approach Enables Accurate Prediction of Reaction Free Energies in Solution DOI Creative Commons
Felix Pultar,

Moritz Thürlemann,

Igor Gordiy

et al.

Journal of the American Chemical Society, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 17, 2025

We present the design and implementation of a novel neural network potential (NNP) its combination with an electrostatic embedding scheme, commonly used within context hybrid quantum-mechanical/molecular-mechanical (QM/MM) simulations. Substitution computationally expensive QM Hamiltonian by NNP same accuracy largely reduces computational cost enables efficient sampling in prospective MD simulations, main limitation faced traditional QM/MM setups. The model relies on recently introduced anisotropic message passing (AMP) formalism to compute atomic interactions encode symmetries found systems. AMP is shown be highly terms both data costs can readily scaled sample systems involving more than 350 solute 40,000 solvent atoms for hundreds nanoseconds using umbrella sampling. Most deviations predictions from underlying DFT ground truth lie chemical (4.184 kJ mol–1). performance broad applicability our approach are showcased calculating free-energy surface alanine dipeptide, preferred ligation states nickel phosphine complexes, dissociation free energies charged pyridine quinoline dimers. Results this ML/MM show excellent agreement experimental reach most cases. In contrast, calculated static calculations paired implicit models or simulations cheaper semiempirical methods up ten times higher deviation sometimes even fail reproduce qualitative trends.

Language: Английский

Citations

0
Accelerating crystal structure search through active learning with neural networks for rapid relaxations DOI Creative Commons
Stefaan S. P. Hessmann, Kristof T. Schütt, Niklas W. A. Gebauer

et al.

npj Computational Materials, Journal Year: 2025, Volume and Issue: 11(1)

Published: Feb. 20, 2025

Language: Английский

Citations

0