Physics-Inspired Structural Representations for Molecules and Materials

Chemical Reviews, Journal Year: 2021, Volume and Issue: 121(16), P. 9759 - 9815

Published: July 26, 2021

The first step in the construction of a regression model or data-driven analysis, aiming to predict elucidate relationship between atomic-scale structure matter and its properties, involves transforming Cartesian coordinates atoms into suitable representation. development representations has played, continues play, central role success machine-learning methods for chemistry materials science. This review summarizes current understanding nature characteristics most commonly used structural chemical descriptions atomistic structures, highlighting deep underlying connections different frameworks ideas that lead computationally efficient universally applicable models. It emphasizes link their physical chemistry, mathematical description, provides examples recent applications diverse set science problems, outlines open questions promising research directions field.

Language: Английский

---

Atomistic Line Graph Neural Network for improved materials property predictions

npj Computational Materials, Journal Year: 2021, Volume and Issue: 7(1)

Published: Nov. 15, 2021

Abstract Graph neural networks (GNN) have been shown to provide substantial performance improvements for atomistic material representation and modeling compared with descriptor-based machine learning models. While most existing GNN models predictions are based on atomic distance information, they do not explicitly incorporate bond angles, which critical distinguishing many structures. Furthermore, properties known be sensitive slight changes in angles. We present an Atomistic Line Neural Network (ALIGNN), a architecture that performs message passing both the interatomic graph its line corresponding demonstrate angle information can efficiently included, leading improved multiple prediction tasks. ALIGNN predicting 52 solid-state molecular available JARVIS-DFT, Materials project, QM9 databases. outperform some previously reported tasks better or comparable model training speed.

Language: Английский

---

Graph neural networks for materials science and chemistry

Communications Materials, Journal Year: 2022, Volume and Issue: 3(1)

Published: Nov. 26, 2022

Abstract Machine learning plays an increasingly important role in many areas of chemistry and materials science, being used to predict properties, accelerate simulations, design new structures, synthesis routes materials. Graph neural networks (GNNs) are one the fastest growing classes machine models. They particular relevance for as they directly work on a graph or structural representation molecules therefore have full access all relevant information required characterize In this Review, we provide overview basic principles GNNs, widely datasets, state-of-the-art architectures, followed by discussion wide range recent applications GNNs concluding with road-map further development application GNNs.

Language: Английский

---

Artificial intelligence in drug discovery: recent advances and future perspectives

Expert Opinion on Drug Discovery, Journal Year: 2021, Volume and Issue: 16(9), P. 949 - 959

Published: March 29, 2021

Introduction: Artificial intelligence (AI) has inspired computer-aided drug discovery. The widespread adoption of machine learning, in particular deep multiple scientific disciplines, and the advances computing hardware software, among other factors, continue to fuel this development. Much initial skepticism regarding applications AI pharmaceutical discovery started vanish, consequently benefitting medicinal chemistry.Areas covered: current status chemoinformatics is reviewed. topics discussed herein include quantitative structure-activity/property relationship structure-based modeling, de novo molecular design, chemical synthesis prediction. Advantages limitations learning are highlighted, together with a perspective on next-generation for discovery.Expert opinion: Deep learning-based approaches have only begun address some fundamental problems Certain methodological advances, such as message-passing models, spatial-symmetry-preserving networks, hybrid innovative paradigms, will likely become commonplace help most challenging questions. Open data sharing model development play central role advancement AI.

Language: Английский

---

SpookyNet: Learning force fields with electronic degrees of freedom and nonlocal effects

Nature Communications, Journal Year: 2021, Volume and Issue: 12(1)

Published: Dec. 14, 2021

Machine-learned force fields (ML-FFs) combine the accuracy of ab initio methods with efficiency conventional fields. However, current ML-FFs typically ignore electronic degrees freedom, such as total charge or spin state, and assume chemical locality, which is problematic when molecules have inconsistent states, nonlocal effects play a significant role. This work introduces SpookyNet, deep neural network for constructing explicit treatment freedom quantum nonlocality. Chemically meaningful inductive biases analytical corrections built into architecture allow it to properly model physical limits. SpookyNet improves upon state-of-the-art (or achieves similar performance) on popular chemistry data sets. Notably, able generalize across conformational space can leverage learned insights, e.g. by predicting unknown thus helping close further important remaining gap today's machine learning models in chemistry.

Language: Английский

---

Benchmarking graph neural networks for materials chemistry

npj Computational Materials, Journal Year: 2021, Volume and Issue: 7(1)

Published: June 3, 2021

Abstract Graph neural networks (GNNs) have received intense interest as a rapidly expanding class of machine learning models remarkably well-suited for materials applications. To date, number successful GNNs been proposed and demonstrated systems ranging from crystal stability to electronic property prediction surface chemistry heterogeneous catalysis. However, consistent benchmark these remains lacking, hindering the development evaluation new in field. Here, we present workflow testing platform, MatDeepLearn, quickly reproducibly assessing comparing other models. We use this platform optimize evaluate selection top performing on several representative datasets computational chemistry. From our investigations note importance hyperparameter find roughly similar performances once optimized. identify strengths over conventional cases with compositionally diverse its overall flexibility respect inputs, due learned rather than defined representations. Meanwhile weaknesses are also observed including high data requirements, suggestions further improvement applications discussed.

Language: Английский

---

Neural Network Potentials: A Concise Overview of Methods

Annual Review of Physical Chemistry, Journal Year: 2022, Volume and Issue: 73(1), P. 163 - 186

Published: Jan. 4, 2022

In the past two decades, machine learning potentials (MLPs) have reached a level of maturity that now enables applications to large-scale atomistic simulations wide range systems in chemistry, physics, and materials science. Different algorithms been used with great success construction these MLPs. this review, we discuss an important group MLPs relying on artificial neural networks establish mapping from atomic structure potential energy. spite common feature, there are conceptual differences among MLPs, which concern dimensionality systems, inclusion long-range electrostatic interactions, global phenomena like nonlocal charge transfer, type descriptor represent structure, can be either predefined or learnable. A concise overview is given along discussion open challenges field.

Language: Английский

---

Machine Learning Methods for Small Data Challenges in Molecular Science

Chemical Reviews, Journal Year: 2023, Volume and Issue: 123(13), P. 8736 - 8780

Published: June 29, 2023

Small data are often used in scientific and engineering research due to the presence of various constraints, such as time, cost, ethics, privacy, security, technical limitations acquisition. However, big have been focus for past decade, small their challenges received little attention, even though they technically more severe machine learning (ML) deep (DL) studies. Overall, challenge is compounded by issues, diversity, imputation, noise, imbalance, high-dimensionality. Fortunately, current era characterized technological breakthroughs ML, DL, artificial intelligence (AI), which enable data-driven discovery, many advanced ML DL technologies developed inadvertently provided solutions problems. As a result, significant progress has made decade. In this review, we summarize analyze several emerging potential molecular science, including chemical biological sciences. We review both basic algorithms, linear regression, logistic regression (LR),

Language: Английский

---

Provably efficient machine learning for quantum many-body problems

Science, Journal Year: 2022, Volume and Issue: 377(6613)

Published: Sept. 22, 2022

Classical machine learning (ML) provides a potentially powerful approach to solving challenging quantum many-body problems in physics and chemistry. However, the advantages of ML over traditional methods have not been firmly established. In this work, we prove that classical algorithms can efficiently predict ground-state properties gapped Hamiltonians after from other same phase matter. By contrast, under widely accepted conjecture, do learn data cannot achieve guarantee. We also classify wide range phases. Extensive numerical experiments corroborate our theoretical results variety scenarios, including Rydberg atom systems, two-dimensional random Heisenberg models, symmetry-protected topological phases, topologically ordered

Language: Английский

---

Perspective on integrating machine learning into computational chemistry and materials science

The Journal of Chemical Physics, Journal Year: 2021, Volume and Issue: 154(23)

Published: June 21, 2021

Machine learning (ML) methods are being used in almost every conceivable area of electronic structure theory and molecular simulation. In particular, ML has become firmly established the construction high-dimensional interatomic potentials. Not a day goes by without another proof principle published on how can represent predict quantum mechanical properties-be they observable, such as polarizabilities, or not, atomic charges. As is becoming pervasive simulation, we provide an overview atomistic computational modeling transformed incorporation approaches. From perspective practitioner field, assess common workflows to structure, dynamics, spectroscopy affected ML. Finally, discuss tighter lasting integration with chemistry materials science be achieved what it will mean for research practice, software development, postgraduate training.

Language: Английский

---