Evaluation of the MACE force field architecture: From medicinal chemistry to materials science

The Journal of Chemical Physics, Journal Year: 2023, Volume and Issue: 159(4)

Published: July 28, 2023

The MACE architecture represents the state of art in field machine learning force fields for a variety in-domain, extrapolation, and low-data regime tasks. In this paper, we further evaluate by fitting models published benchmark datasets. We show that generally outperforms alternatives wide range systems, from amorphous carbon, universal materials modeling, general small molecule organic chemistry to large molecules liquid water. demonstrate capabilities model on tasks ranging constrained geometry optimization molecular dynamics simulations find excellent performance across all tested domains. is very data efficient can reproduce experimental vibrational spectra when trained as few 50 randomly selected reference configurations. strictly local atom-centered sufficient such even case weakly interacting assemblies.

Language: Английский

---

SchNetPack 2.0: A neural network toolbox for atomistic machine learning

The Journal of Chemical Physics, Journal Year: 2023, Volume and Issue: 158(14)

Published: March 21, 2023

SchNetPack is a versatile neural network toolbox that addresses both the requirements of method development and application atomistic machine learning. Version 2.0 comes with an improved data pipeline, modules for equivariant networks, PyTorch implementation molecular dynamics. An optional integration Lightning Hydra configuration framework powers flexible command-line interface. This makes easily extendable custom code ready complex training tasks, such as generation 3D structures.

Language: Английский

---

Biomolecular dynamics with machine-learned quantum-mechanical force fields trained on diverse chemical fragments

Science Advances, Journal Year: 2024, Volume and Issue: 10(14)

Published: April 5, 2024

The GEMS method enables molecular dynamics simulations of large heterogeneous systems at ab initio quality.

Language: Английский

---

Enhancing geometric representations for molecules with equivariant vector-scalar interactive message passing

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: Jan. 5, 2024

Abstract Geometric deep learning has been revolutionizing the molecular modeling field. Despite state-of-the-art neural network models are approaching ab initio accuracy for property prediction, their applications, such as drug discovery and dynamics (MD) simulation, have hindered by insufficient utilization of geometric information high computational costs. Here we propose an equivariant geometry-enhanced graph called ViSNet, which elegantly extracts features efficiently structures with low Our proposed ViSNet outperforms approaches on multiple MD benchmarks, including MD17, revised MD17 MD22, achieves excellent chemical prediction QM9 Molecule3D datasets. Furthermore, through a series simulations case studies, can explore conformational space provide reasonable interpretability to map representations structures.

Language: Английский

---

Navigating the molecular landscape of environmental science and heavy metal removal: A simulation-based approach

Journal of Molecular Liquids, Journal Year: 2024, Volume and Issue: 410, P. 125592 - 125592

Published: July 20, 2024

Heavy metals pose a significant threat to ecosystems and human health because of their toxic properties ability bioaccumulate in living organisms. Traditional removal methods often fall short terms cost, energy efficiency, minimizing secondary pollutant generation, especially complex environmental settings. In contrast, molecular simulation offer promising solution by providing in-depth insights into atomic interactions between heavy potential adsorbents. This review highlights the for removing types pollutants science, specifically metals. These powerful tool predicting designing materials processes remediation. We focus on specific like lead, Cadmium, mercury, utilizing cutting-edge techniques such as Molecular Dynamics (MD), Monte Carlo (MC) simulations, Quantum Chemical Calculations (QCC), Artificial Intelligence (AI). By leveraging these methods, we aim develop highly efficient selective unravelling underlying mechanisms, pave way developing more technologies. comprehensive addresses critical gap scientific literature, valuable researchers protection health. modelling hold promise revolutionizing prediction metals, ultimately contributing sustainable solutions cleaner healthier future.

Language: Английский

---

A Euclidean transformer for fast and stable machine learned force fields

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: Aug. 6, 2024

Abstract Recent years have seen vast progress in the development of machine learned force fields (MLFFs) based on ab-initio reference calculations. Despite achieving low test errors, reliability MLFFs molecular dynamics (MD) simulations is facing growing scrutiny due to concerns about instability over extended simulation timescales. Our findings suggest a potential connection between robustness cumulative inaccuracies and use equivariant representations MLFFs, but computational cost associated with these can limit this advantage practice. To address this, we propose transformer architecture called SO3krates that combines sparse ( Euclidean variables ) self-attention mechanism separates invariant information, eliminating need for expensive tensor products. achieves unique combination accuracy, stability, speed enables insightful analysis quantum properties matter time system size scales. showcase capability, generate stable MD trajectories flexible peptides supra-molecular structures hundreds atoms. Furthermore, investigate PES topology medium-sized chainlike molecules (e.g., small peptides) by exploring thousands minima. Remarkably, demonstrates ability strike balance conflicting demands stability emergence new minimum-energy conformations beyond training data, which crucial realistic exploration tasks field biochemistry.

Language: Английский

---

Recent Advances in Machine Learning‐Assisted Multiscale Design of Energy Materials

Advanced Energy Materials, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 10, 2024

Abstract This review highlights recent advances in machine learning (ML)‐assisted design of energy materials. Initially, ML algorithms were successfully applied to screen materials databases by establishing complex relationships between atomic structures and their resulting properties, thus accelerating the identification candidates with desirable properties. Recently, development highly accurate interatomic potentials generative models has not only improved robust prediction physical but also significantly accelerated discovery In past couple years, methods have enabled high‐precision first‐principles predictions electronic optical properties for large systems, providing unprecedented opportunities science. Furthermore, ML‐assisted microstructure reconstruction physics‐informed solutions partial differential equations facilitated understanding microstructure–property relationships. Most recently, seamless integration various platforms led emergence autonomous laboratories that combine quantum mechanical calculations, language models, experimental validations, fundamentally transforming traditional approach novel synthesis. While highlighting aforementioned advances, existing challenges are discussed. Ultimately, is expected fully integrate atomic‐scale simulations, reverse engineering, process optimization, device fabrication, empowering system design. will drive transformative innovations conversion, storage, harvesting technologies.

Language: Английский

---

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

Journal of Chemical Theory and Computation, Journal Year: 2024, Volume and Issue: 20(10), P. 4076 - 4087

Published: May 14, 2024

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

Language: Английский

---

Crash testing machine learning force fields for molecules, materials, and interfaces: molecular dynamics in the TEA challenge 2023

Chemical Science, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 1, 2025

We present a comprehensive analysis of the capabilities modern machine learning force fields to simulate long-term molecular dynamics at near-ambient conditions for molecules, molecule-surface interfaces, and materials within TEA Challenge 2023.

Language: Английский

---

Crash testing machine learning force fields for molecules, materials, and interfaces: model analysis in the TEA Challenge 2023

Chemical Science, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 1, 2025

Assessing the performance of modern machine learning force fields across diverse chemical systems to identify their strengths and limitations within TEA Challenge 2023.

Language: Английский

---