Recent advances and applications of deep learning methods in materials science

npj Computational Materials, Journal Year: 2022, Volume and Issue: 8(1)

Published: April 5, 2022

Deep learning (DL) is one of the fastest growing topics in materials data science, with rapidly emerging applications spanning atomistic, image-based, spectral, and textual modalities. DL allows analysis unstructured automated identification features. Recent development large databases has fueled application methods atomistic prediction particular. In contrast, advances image spectral have largely leveraged synthetic enabled by high quality forward models as well generative unsupervised methods. this article, we present a high-level overview deep-learning followed detailed discussion recent developments deep simulation, imaging, analysis, natural language processing. For each modality discuss involving both theoretical experimental data, typical modeling approaches their strengths limitations, relevant publicly available software datasets. We conclude review cross-cutting work related to uncertainty quantification field brief perspective on challenges, potential growth areas for science. The science presents an exciting avenue future discovery design.

Language: Английский

---

Machine learning for high-entropy alloys: Progress, challenges and opportunities

Progress in Materials Science, Journal Year: 2022, Volume and Issue: 131, P. 101018 - 101018

Published: Sept. 15, 2022

Language: Английский

---

High-throughput predictions of metal–organic framework electronic properties: theoretical challenges, graph neural networks, and data exploration

npj Computational Materials, Journal Year: 2022, Volume and Issue: 8(1)

Published: May 17, 2022

Abstract With the goal of accelerating design and discovery metal–organic frameworks (MOFs) for electronic, optoelectronic, energy storage applications, we present a dataset predicted electronic structure properties thousands MOFs carried out using multiple density functional approximations. Compared to more accurate hybrid functionals, find that widely used PBE generalized gradient approximation (GGA) severely underpredicts MOF band gaps in largely systematic manner semi-conductors insulators without magnetic character. However, an even larger less predictable disparity gap prediction is with open-shell 3 d transition metal cations. regards partial atomic charges, different approximations predict similar charges overall, although functionals tend shift electron away from centers onto ligand environments compared GGA point reference. Much significant differences are observed when comparing charge partitioning schemes. We conclude by computed train machine-learning models can rapidly all four considered this work, paving way future high-throughput screening studies. To encourage exploration reuse theoretical calculations presented curated data made publicly available via interactive user-friendly web application on Materials Project.

Language: Английский

---

Human- and machine-centred designs of molecules and materials for sustainability and decarbonization

Nature Reviews Materials, Journal Year: 2022, Volume and Issue: 7(12), P. 991 - 1009

Published: Aug. 24, 2022

Language: Английский

---

MOFormer: Self-Supervised Transformer Model for Metal–Organic Framework Property Prediction

Journal of the American Chemical Society, Journal Year: 2023, Volume and Issue: 145(5), P. 2958 - 2967

Published: Jan. 27, 2023

Metal-organic frameworks (MOFs) are materials with a high degree of porosity that can be used for many applications. However, the chemical space MOFs is enormous due to large variety possible combinations building blocks and topology. Discovering optimal specific applications requires an efficient accurate search over countless potential candidates. Previous high-throughput screening methods using computational simulations like DFT time-consuming. Such also require 3D atomic structures MOFs, which adds one extra step when evaluating hypothetical MOFs. In this work, we propose structure-agnostic deep learning method based on Transformer model, named as MOFormer, property predictions MOFormer takes text string representation MOF (MOFid) input, thus circumventing need obtaining structure accelerating process. By comparing other descriptors such Stoichiometric-120 revised autocorrelations, demonstrate achieve state-of-the-art prediction accuracy all benchmarks. Furthermore, introduce self-supervised framework pretrains via maximizing cross-correlation between its representations structure-based crystal graph convolutional neural network (CGCNN) >400k publicly available data. Benchmarks show pretraining improves both models various downstream tasks. revealed more data-efficient quantum-chemical than CGCNN training data limited. Overall, provides novel perspective learning.

Language: Английский

---

Representations of Materials for Machine Learning

Annual Review of Materials Research, Journal Year: 2023, Volume and Issue: 53(1), P. 399 - 426

Published: April 18, 2023

High-throughput data generation methods and machine learning (ML) algorithms have given rise to a new era of computational materials science by the relations between composition, structure, properties exploiting such for design. However, build these connections, must be translated into numerical form, called representation, that can processed an ML model. Data sets in vary format (ranging from images spectra), size, fidelity. Predictive models scope interest. Here, we review context-dependent strategies constructing representations enable use as inputs or outputs models. Furthermore, discuss how modern techniques learn transfer chemical physical information tasks. Finally, outline high-impact questions not been fully resolved thus require further investigation.

Language: Английский

---

Revolutionizing the structural design and determination of covalent–organic frameworks: principles, methods, and techniques

Chemical Society Reviews, Journal Year: 2023, Volume and Issue: 53(1), P. 502 - 544

Published: Dec. 15, 2023

Covalent organic frameworks (COFs) represent an important class of crystalline porous materials with designable structures and functions. The interconnected monomers, featuring pre-designed symmetries connectivities, dictate the COFs, endowing them high thermal chemical stability, large surface area, tunable micropores. Furthermore, by utilizing pre-functionalization or post-synthetic functionalization strategies, COFs can acquire multifunctionalities, leading to their versatile applications in gas separation/storage, catalysis, optoelectronic devices. Our review provides a comprehensive account latest advancements principles, methods, techniques for structural design determination COFs. These cutting-edge approaches enable rational precise elucidation COF structures, addressing fundamental physicochemical challenges associated host-guest interactions, topological transformations, network interpenetration, defect-mediated catalysis.

Language: Английский

---

Updates to the DScribe library: New descriptors and derivatives

The Journal of Chemical Physics, Journal Year: 2023, Volume and Issue: 158(23)

Published: June 20, 2023

We present an update of the DScribe package, a Python library for atomistic descriptors. The extends DScribe's descriptor selection with Valle-Oganov materials fingerprint and provides derivatives to enable more advanced machine learning tasks, such as force prediction structure optimization. For all descriptors, numeric are now available in DScribe. many-body tensor representation (MBTR) Smooth Overlap Atomic Positions (SOAP), we have also implemented analytic derivatives. demonstrate effectiveness models Cu clusters perovskite alloys.

Language: Английский

---

Extrapolative prediction of small-data molecular property using quantum mechanics-assisted machine learning

npj Computational Materials, Journal Year: 2024, Volume and Issue: 10(1)

Published: Jan. 10, 2024

Abstract Data-driven materials science has realized a new paradigm by integrating domain knowledge and machine-learning (ML) techniques. However, ML-based research often overlooked the inherent limitation in predicting unknown data: extrapolative performance, especially when dealing with small-scale experimental datasets. Here, we present comprehensive benchmark for assessing performance across 12 organic molecular properties. Our large-scale reveals that conventional ML models exhibit remarkable degradation beyond training distribution of property range structures, particularly small-data To address this challenge, introduce quantum-mechanical (QM) descriptor dataset, called QMex, an interactive linear regression (ILR), which incorporates interaction terms between QM descriptors categorical information pertaining to structures. The QMex-based ILR achieved state-of-the-art while preserving its interpretability. results, QMex proposed model serve as valuable assets improving predictions small datasets discovery novel materials/molecules surpass existing candidates.

Language: Английский

---

GNNs for mechanical properties prediction of strut-based lattice structures

International Journal of Mechanical Sciences, Journal Year: 2024, Volume and Issue: 269, P. 109082 - 109082

Published: Feb. 2, 2024

Language: Английский

---