Data Generation for Machine Learning Interatomic Potentials and Beyond

Chemical Reviews, Journal Year: 2024, Volume and Issue: 124(24), P. 13681 - 13714

Published: Nov. 21, 2024

The field of data-driven chemistry is undergoing an evolution, driven by innovations in machine learning models for predicting molecular properties and behavior. Recent strides ML-based interatomic potentials have paved the way accurate modeling diverse chemical structural at atomic level. key determinant defining MLIP reliability remains quality training data. A paramount challenge lies constructing sets that capture specific domains vast space. This Review navigates intricate landscape essential components integrity data ensure extensibility transferability resulting models. We delve into details active learning, discussing its various facets implementations. outline different types uncertainty quantification applied to atomistic acquisition correlations between estimated true error. role samplers generating informative structures highlighted. Furthermore, we discuss via modified surrogate potential energy surfaces as innovative approach diversify also provides a list publicly available cover

Language: Английский

---

Modern semiempirical electronic structure methods and machine learning potentials for drug discovery: Conformers, tautomers, and protonation states

The Journal of Chemical Physics, Journal Year: 2023, Volume and Issue: 158(12)

Published: March 6, 2023

Modern semiempirical electronic structure methods have considerable promise in drug discovery as universal “force fields” that can reliably model biological and drug-like molecules, including alternative tautomers protonation states. Herein, we compare the performance of several neglect diatomic differential overlap-based (MNDO/d, AM1, PM6, PM6-D3H4X, PM7, ODM2), density-functional tight-binding based (DFTB3, DFTB/ChIMES, GFN1-xTB, GFN2-xTB) models with pure machine learning potentials (ANI-1x ANI-2x) hybrid quantum mechanical/machine (AIQM1 QDπ) for a wide range data computed at consistent ωB97X/6-31G* level theory (as ANI-1x database). This includes conformational energies, intermolecular interactions, tautomers, Additional comparisons are made to set natural synthetic nucleic acids from artificially expanded genetic information system has important implications design new biotechnology therapeutics. Finally, examine acid/base chemistry relevant RNA cleavage reactions catalyzed by small nucleolytic ribozymes, DNAzymes, ribonucleases. Overall, appear be most robust these datasets, recently developed QDπ performs exceptionally well, having especially high accuracy states discovery.

Language: Английский

---

AI in computational chemistry through the lens of a decade-long journey

Chemical Communications, Journal Year: 2024, Volume and Issue: 60(24), P. 3240 - 3258

Published: Jan. 1, 2024

This article gives a perspective on the progress of AI tools in computational chemistry through lens author's decade-long contributions put wider context trends this rapidly expanding field. over last decade is tremendous: while ago we had glimpse what was to come many proof-of-concept studies, now witness emergence AI-based that are mature enough make faster and more accurate simulations increasingly routine. Such turn allow us validate even revise experimental results, deepen our understanding physicochemical processes nature, design better materials, devices, drugs. The rapid introduction powerful rise unique challenges opportunities discussed too.

Language: Английский

---

Modern semiempirical electronic structure methods

The Journal of Chemical Physics, Journal Year: 2024, Volume and Issue: 160(4)

Published: Jan. 24, 2024

---

Constructing Accurate and Efficient General-Purpose Atomistic Machine Learning Model with Transferable Accuracy for Quantum Chemistry

Journal of Chemical Theory and Computation, Journal Year: 2024, Volume and Issue: 20(21), P. 9500 - 9511

Published: Oct. 31, 2024

Density functional theory (DFT) has been a cornerstone in computational science, providing powerful insights into structure-property relationships for molecules and materials through first-principles quantum-mechanical (QM) calculations. However, the advent of atomistic machine learning (ML) is reshaping landscape by enabling large-scale dynamics simulations high-throughput screening at DFT-equivalent accuracy with drastically reduced cost. Yet, development general-purpose ML models as surrogates QM calculations faces several challenges, particularly terms model capacity, data efficiency, transferability across chemically diverse systems. This work introduces novel extension polarizable atom interaction neural network (namely, XPaiNN) to address these challenges. Two distinct training strategies have employed, one direct-learning other Δ-ML on top semiempirical method. These methodologies implemented within same framework, allowing detailed comparison their results. The XPaiNN models, particular using Δ-ML, not only demonstrate competitive performance standard benchmarks, but also effectiveness against methods comprehensive downstream tasks, including noncovalent interactions, reaction energetics, barrier heights, geometry optimization thermodynamics, etc. represents significant step forward pursuit accurate efficient general-purpose, capable handling complex chemical systems transferable accuracy.

Language: Английский

---

Universal and Updatable Artificial Intelligence-Enhanced Quantum Chemical Foundational Models

Published: June 26, 2024

Quantum chemical methods developed since 1927 are instrumental in simulations but human expertise has been still essential choosing a suitable method. Here we introduce paradigm shift to universal and updatable artificial intelligence-enhanced quantum mechanical (UAIQM) foundational models with an online platform auto-selecting the best accuracy for given system, available time, moderate computational resources (see https://xacs.xmu.edu.cn/docs/mlatom/tutorial_uaiqm.html instructions). The hosts growing library of state-of-the-art UAIQM calibrated uncertainties provides mechanism improving continuously more usage. We demonstrate how can be used massive accurate within hours on commodity hardware which would take days or weeks high-performance computing centers less workhorse methods. also show that sets new standard infrared spectra, reaction barriers, energetics whose predictions have far-reaching consequences molecular simulations.

Language: Английский

---

Pairwise Difference Learning for Classification

Lecture notes in computer science, Journal Year: 2025, Volume and Issue: unknown, P. 284 - 299

Published: Jan. 1, 2025

Language: Английский

---

Kernel regression methods for prediction of materials properties: Recent developments

Chemical Physics Reviews, Journal Year: 2025, Volume and Issue: 6(1)

Published: Feb. 13, 2025

Machine learning (ML) is increasingly used in chemical physics and materials science. One major area of thrust machine properties molecules solid from descriptors composition structure. Recently, kernel regression methods various flavors—such as ridge regression, Gaussian process support vector machine—have attracted attention such applications. Kernel allow benefiting simultaneously the advantages linear regressions superior expressive power nonlinear kernels. In many applications, are high-dimensional feature spaces, where sampling with training data bound to be sparse effects specific spaces significantly affect performance method. We review recent applications kernel-based for prediction structure related purposes. discuss methodological aspects including choices kernels appropriate different dimensionality, ways balance reliability model data. also regression-based hybrid ML approaches.

Language: Английский

---

Facilitating Transition State Search with Minimal Conformational Sampling Using Reaction Graph

Journal of Chemical Theory and Computation, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 25, 2025

Elucidating transition states (TSs) is crucial for understanding chemical reactions. The reliability of traditional TS search approaches depends on input conformations that require significant effort to prepare. Previous automated methods generating reaction typically involve extensive exploration a large conformational space. Such exhaustive can be complicated by the rapid growth space, especially reactions involving many rotatable bonds, multiple reacting molecules, and numerous bond formations dissociations. To address this problem, we propose new approach generates searches with minimal reliance sampling. This method constructs pseudo-TS structure based graph containing formation dissociation information modifies it produce reactant product conformations. Tested three different benchmarks, our consistently generated suitable without necessitating sampling, demonstrating its potential significantly improve applicability searches. offers valuable tool broad range applications such as mechanism analysis network exploration.

Language: Английский

---

Advancements in Machine Learning Predicting Activation and Gibbs Free Energies in Chemical Reactions

International Journal of Quantum Chemistry, Journal Year: 2025, Volume and Issue: 125(7)

Published: March 19, 2025

ABSTRACT Machine learning has revolutionized computational chemistry by improving the accuracy of predicting thermodynamic and kinetic properties like activation energies Gibbs free energies, accelerating materials discovery optimizing reaction conditions in both academic industrial applications. This review investigates recent strides applying advanced machine techniques, including transfer learning, for accurately within complex chemical reactions. It thoroughly provides an extensive overview pivotal methods utilized this domain, sophisticated neural networks, Gaussian processes, symbolic regression. Furthermore, prominently highlights commonly adopted frameworks, such as Chemprop, SchNet, DeepMD, which have consistently demonstrated remarkable exceptional efficiency properties. Moreover, it carefully explores numerous influential studies that notably reported substantial successes, particularly focusing on predictive performance, diverse datasets, innovative model architectures profoundly contributed to enhancing methodologies. Ultimately, clearly underscores transformative potential significantly power intricate systems, bearing considerable implications cutting‐edge theoretical research practical

Language: Английский

---