Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties

Artificial Intelligence and High-Throughput Computational Workflows Empowering the Fast Screening of Metal-Organic Frameworks for Hydrogen Storage

Metal-organic frameworks (MOFs) are one of the most promising hydrogen-storing materials due to their rich specific surface area, adjustable topological and pore structures, and modified functional groups. In this work, we developed automatically parallel computational workflows for high-throughput screening of ∼11,600 MOFs from the CoRE database and discovered 69 top-performing MOF candidates with work capacity greater than 1.00 wt % at 298.5 K and a pressure swing between 100 and 0.1 bar, which is at least twice that of MOF-5. In particular, ZITRUP, OQFAJ01, WANHOL, and VATYIZ showed excellent hydrogen storage performance of 4.48, 3.16, 2.19, and 2.16 wt %. We specifically analyzed the relationship between pore-limiting diameter, largest cavity diameter, void fraction, open metal sites, metal elements or nonmetallic atomic elements, and deliverable capacity and found that not only geometrical and physical features of crystalline but also chemical properties of adsorbate sites determined the H2 storage capacity of MOFs at room temperature. It is highlighted that we first proposed the modified crystal graph convolutional neural networks by incorporating the obtained geometrical and physical features into the convolutional high-dimensional feature vectors of period crystal structures for predicting H2 storage performance, which can improve the prediction accuracy of the neural network from the former mean absolute error (MAE) of 0.064 wt % to the current MAE of 0.047 wt % and shorten the consuming time to about 10-4 times of high-throughput computational screening. This work opens a new avenue toward high-throughput screening of MOFs for H2 adsorption capacity, which can be extended for the screening and discovery of other functional materials.

AI-assisted chemistry research: a comprehensive analysis of evolutionary paths and hotspots through knowledge graphs

Artificial intelligence (AI) offers transformative potential for chemical research through its ability to optimize reactions and processes, enhance energy efficiency, and reduce waste. AI-assisted chemical research (AI + chem) has become a global hotspot. To better understand the current research status of "AI + chem", this study conducted a scientific bibliometric investigation using CiteSpace. The web of science core collection was utilized to retrieve original articles related to "AI + chem" published from 2000 to 2024. The obtained data allowed for the visualization of the knowledge background, current research status, and latest knowledge structure of "AI + chem". The "AI + chem" has entered a stage of explosive growth, and the number of papers will maintain long-term high-speed growth. This article systematically analyzes the latest progress in "AI + chem" and objectively predicts future trends, including molecular design, reaction prediction, materials design, drug design, and quantum chemistry. The outcomes of this study will provide readers with a comprehensive understanding of the overall landscape of "AI + chem".

Machine Learning Classification of Local Environments in Molecular Crystals

Identifying local structural motifs and packing patterns of molecular solids is a challenging task for both simulation and experiment. We demonstrate two novel approaches to characterize local environments in different polymorphs of molecular crystals using learning models that employ either flexibly learned or handcrafted molecular representations. In the first case, we follow our earlier work on graph learning in molecular crystals, deploying an atomistic graph convolutional network combined with molecule-wise aggregation to enable per-molecule environmental classification. For the second model, we develop a new set of descriptors based on symmetry functions combined with a point-vector representation of the molecules, encoding information about the positions and relative orientations of the molecule. We demonstrate very high classification accuracy for both approaches on urea and nicotinamide crystal polymorphs and practical applications to the analysis of dynamical trajectory data for nanocrystals and solid-solid interfaces. Both architectures are applicable to a wide range of molecules and diverse topologies, providing an essential step in the exploration of complex condensed matter phenomena.

AtomGPT: Atomistic Generative Pretrained Transformer for Forward and Inverse Materials Design

Large language models (LLMs) such as generative pretrained transformers (GPTs) have shown potential for various commercial applications, but their applicability for materials design remains underexplored. In this Letter, AtomGPT is introduced as a model specifically developed for materials design based on transformer architectures, demonstrating capabilities for both atomistic property prediction and structure generation. This study shows that a combination of chemical and structural text descriptions can efficiently predict material properties with accuracy comparable to graph neural network models, including formation energies, electronic bandgaps from two different methods, and superconducting transition temperatures. Furthermore, AtomGPT can generate atomic structures for tasks such as designing new superconductors, with the predictions validated through density functional theory calculations. This work paves the way for leveraging LLMs in forward and inverse materials design, offering an efficient approach to the discovery and optimization of materials.

Self-assembly prediction of architecture-controlled bottlebrush copolymers in solution using graph convolutional networks

The investigation of bottlebrush copolymer self-assembly in solution involves a comprehensive approach integrating simulation and experimental research, due to their unique physical characteristics. However, the intricate architecture of bottlebrush copolymers and the diverse solvent conditions introduce a wide range of parameter spaces. In this study, we investigated the solution self-assembly behavior of bottlebrush copolymers by combining dissipative particle dynamics (DPD) simulation results and machine learning (ML) including graph convolutional networks (GCNs). The architecture of bottlebrush copolymers is encoded by graphs including connectivity, side chain length, bead types, and interaction parameters of DPD simulation. Using GCN, we accurately predicted the single chain properties of bottlebrush copolymers with over 95% accuracy. Furthermore, phase behavior was precisely predicted using these single chain properties. Shapley additive explanations (SHAP) values of single chain properties to the various self-assembly morphologies were calculated to investigate the correlation between single chain properties and morphologies. In addition, we analyzed single chain properties and phase behavior as a function of DPD interaction parameters, extracting relevant physical properties for vesicle morphology formation. This work paves the way for tailored design in solution of self-assembled nanostructures of bottlebrush copolymers, offering a GCN framework for precise prediction of self-assembly morphologies under various chain architectures and solvent conditions.

Classifying and Predicting the Thermal Expansion Properties of Metal-Organic Frameworks: A Data-Driven Approach

Metal-organic frameworks (MOFs) are versatile materials for a wide variety of potential applications. Tunable thermal expansion properties promote the application of MOFs in thermally sensitive composite materials; however, they are currently available only in a handful of structures. Herein, we report the first data set for thermal expansion properties of 33,131 diverse MOFs generated from molecular simulations and subsequently develop machine learning (ML) models to (1) classify different thermal expansion behaviors and (2) predict volumetric thermal expansion coefficients (αV). The random forest model trained on hybrid descriptors combining geometric, chemical, and topological features exhibits the best performance among different ML models. Based on feature importance analysis, linker chemistry and topological arrangement are revealed to have a dominant impact on thermal expansion. Furthermore, we identify common building blocks in MOFs with exceptional thermal expansion properties. This data-driven study is the first of its kind, not only constructing a useful data set to facilitate future studies on this important topic but also providing design guidelines for advancing new MOFs with desired thermal expansion properties.

PACMAN: A Robust Partial Atomic Charge Predicter for Nanoporous Materials Based on Crystal Graph Convolution Networks

We report a fast and easy method (PACMAN) to assign partial atomic charges on metal-organic framework (MOF) and covalent-organic framework (COF) crystal structures based on graph convolution networks (GCNs) trained on >1.8 million high-fidelity partial atomic charge data obtained from the Quantum Metal-Organic Framework (QMOF) database. The developed model shows outstanding performance, achieving a mean absolute error (MAE) of 0.0055 e (test set performance) while maintaining consistency with DDEC6, Bader, and CM5 charges across diverse chemistry and topologies of MOFs and COFs. We find that the new method accurately assigns partial atomic charges for ion-containing nanoporous materials, which has not been possible in previous machine learning (ML) models. Grand canonical Monte Carlo (GCMC) simulation results for CO2 and N2 uptakes and the Widom particle insertion calculation for Henry's law constant of water results based on PACMAN and the original DDEC6 charges show excellent agreements compared to other ML models reported in the literature. The runtime analysis of the new method demonstrates that the partial atomic charges of MOF and COF structures with up to 500 atoms can be obtained in less than 10 s. An easy-to-use web interface has been developed to facilitate the adoption of the developed model.

Addressing the Accuracy-Cost Trade-off in Material Property Prediction Using a Teacher-Student Strategy

Deep learning has catalyzed a transformative shift in material discovery, offering a key advantage over traditional experimental and theoretical methods by significantly reducing associated costs. Models adept at predicting properties from chemical compositions alone do not require structural information. However, this cost-efficient approach compromises model precision, particularly in Chemical Composition-based Property Prediction Models (CPMs), which are notably less accurate than Structure-based Property Prediction Models (SPMs). Addressing this challenge, our study introduces a novel Teacher-Student (TS) strategy, where a pretrained SPM serves as an instructive 'teacher' to enhance the CPM's precision. This TS strategy successfully harmonizes low-cost exploration with high accuracy, achieving a significant 47.1% reduction in relative error in scenarios involving 100 data entries. We also evaluate the effectiveness of the proposed strategy by employing perovskites as a case study. This method represents a significant advancement in the exploration and identification of valuable materials, leveraging CPM's potential while overcoming its precision limitations.

Accelerating Molecular Vibrational Spectra Simulations with a Physically Informed Deep Learning Model

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

Application of Transformers in Cheminformatics

By accelerating time-consuming processes with high efficiency, computing has become an essential part of many modern chemical pipelines. Machine learning is a class of computing methods that can discover patterns within chemical data and utilize this knowledge for a wide variety of downstream tasks, such as property prediction or substance generation. The complex and diverse chemical space requires complex machine learning architectures with great learning power. Recently, learning models based on transformer architectures have revolutionized multiple domains of machine learning, including natural language processing and computer vision. Naturally, there have been ongoing endeavors in adopting these techniques to the chemical domain, resulting in a surge of publications within a short period. The diversity of chemical structures, use cases, and learning models necessitate a comprehensive summarization of existing works. In this paper, we review recent innovations in adapting transformers to solve learning problems in chemistry. Because chemical data is diverse and complex, we structure our discussion based on chemical representations. Specifically, we highlight the strengths and weaknesses of each representation, the current progress of adapting transformer architectures, and future directions.

Analyzing the Active Site and Predicting the Overall Activity of Alloy Catalysts

As one of the potential catalysts, disordered solid solution alloys can offer a wealth of catalytic sites. However, accurately evaluating their activity localization structure and overall activity from each individual site remains a formidable challenge. Herein, an approach based on density functional theory and machine learning was used to obtain a large number of sites of the Pt-Ru alloy as the model multisite catalyst for the hydrogen evolution reaction. Subsequently, a series of statistical approaches were employed to unveil the relationship between the geometric structure and overall activity. Based on the radial frequency distribution of metal elements and the distribution of ΔGH, we have identified the surface and subsurface sites occupied by Pt and Ru, respectively, as the most active sites. Particularly, the concept of equivalent site ratio predicts that the overall activity is highest when the Ru content is 20-30%. Furthermore, a series of Pt-Ru alloys were synthesized to validate the proposed theory. This provides crucial insights into understanding the origin of catalytic activity in alloys and thus will better guide the rational development of targeted multisite catalysts.

Machine Learning to Promote Efficient Screening of Low-Contact Electrode for 2D Semiconductor Transistor Under Limited Data

Low-barrier and high-injection electrodes are crucial for high-performance (HP) 2D semiconductor devices. Conventional trial-and-error methodologies for electrode material screening are impractical because of their low efficiency and arbitrary specificity. Although machine learning has emerged as a promising alternative to tackle this problem, its practical application in semiconductor devices is hindered by its substantial data requirements. In this paper, a comprehensive scheme combining an autoencoding regularized adversarial neural network and a feature-adaptive variational active learning algorithm for screening low-contact electrode materials for 2D semiconductor transistors with limited data is proposed. The proposed scheme exhibits exceptional performance by training with only 15% of the total data points, where the mean square errors are 0.17 and 0.27 eV for the vertical and lateral Schottky barrier, respectively, and 2.88% for tunneling probability. Further, it exhibits an optimal predictive performance for 100 randomly sampled training datasets, reveals the underlying physical insight based on the identified features, and realizes continual improvement by employing detailed density-of-states descriptors. Finally, the empirical evaluations of the transport characteristics are conducted and verified by constructing MOSFET devices. These findings demonstrate the considerable potential of machine-learning techniques for screening high-efficiency electrode materials and constructing HP 2D semiconductor devices.

Atom-ProteinQA: Atom-level protein model quality assessment through fine-grained joint learning

MOTIVATION

Protein model quality assessment (ProteinQA) is a fundamental task that is essential for biologically relevant applications, i.e., protein structure refinement, protein design, etc. Previous works aimed to conduct ProteinQA only on the global structure or per-residue level, ignoring potentially usable and precise cues from a fine-grained per-atom perspective. In this study, we propose an atom-level ProteinQA model, named Atom-ProteinQA, in which two innovative modules are designed to extract geometric and topological atom-level relationships respectively. Specifically, on the one hand, a geometric perception module exploits 3D sparse convolution to capture the geometric features of the input protein, generating fine-grained atom-level predictions. On the other hand, natural chemical bonds are utilized to construct an atom-level graph, then message passing from a topological perception module is applied to output residue-level predictions in parallel. Eventually, through a cross-model aggregation module, features from different modules mutually interact, enhancing performance on both the atom and residue levels.

RESULTS

Extensive experiments show that our proposed Atom-ProteinQA outperforms previous methods by a large margin, regardless of residue-level or atom-level assessment. Concretely, we achieved state-of-the-art performance on CATH-2084, Decoy-8000, public benchmarks CASP13 & CASP14, and the CAMEO.

AVAILABILITY

The repository of this project is released on: https://github.com/luyfcandy/Atom_ProteinQA.

In Silico Chemical Experiments in the Age of AI: From Quantum Chemistry to Machine Learning and Back

Computational chemistry is an indispensable tool for understanding molecules and predicting chemical properties. However, traditional computational methods face significant challenges due to the difficulty of solving the Schrödinger equations and the increasing computational cost with the size of the molecular system. In response, there has been a surge of interest in leveraging artificial intelligence (AI) and machine learning (ML) techniques to in silico experiments. Integrating AI and ML into computational chemistry increases the scalability and speed of the exploration of chemical space. However, challenges remain, particularly regarding the reproducibility and transferability of ML models. This review highlights the evolution of ML in learning from, complementing, or replacing traditional computational chemistry for energy and property predictions. Starting from models trained entirely on numerical data, a journey set forth toward the ideal model incorporating or learning the physical laws of quantum mechanics. This paper also reviews existing computational methods and ML models and their intertwining, outlines a roadmap for future research, and identifies areas for improvement and innovation. Ultimately, the goal is to develop AI architectures capable of predicting accurate and transferable solutions to the Schrödinger equation, thereby revolutionizing in silico experiments within chemistry and materials science.

High Glass Transition Temperature Fluorinated Polymers Based on Transfer Learning with Small Experimental Data

Machine learning can be used to predict the properties of polymers and explore vast chemical spaces. However, the limited number of available experimental datasets hinders the enhancement of the predictive performance of a model. This study proposes a machine learning approach that leverages transfer learning and ensemble modeling to efficiently predict the glass transition temperature (Tg) of fluorinated polymers and guide the design of high Tg copolymers. Initially, the quantum machine 9 (QM9) dataset is employed for model pretraining, thus providing robust molecular representations for the subsequent fine-tuning of a specialized copolymer dataset. Ensemble modeling is used to further enhance prediction robustness and reliability, effectively addressing the problems owing to the limited and unevenly distributed nature of the copolymer dataset. Finally, a fine-tuned ensemble model is used to navigate a vast chemical space comprising 61 monomers and identify promising candidates for high Tg fluorinated polymers. The model predicts 247 entries capable of achieving a Tg over 390 K, of which 14 are experimentally validated. This study demonstrates the potential of machine learning in material design and discovery, highlighting the effectiveness of transfer learning and ensemble modeling strategies for overcoming the challenges posed by small datasets in complex copolymer systems.

Merits and Demerits of Machine Learning of Ferroelectric, Flexoelectric, and Electrolytic Properties of Ceramic Materials

In the present review, the merits and demerits of machine learning (ML) in materials science are discussed, compared with first principles calculations (PDE (partial differential equations) model) and physical or phenomenological ODE (ordinary differential equations) model calculations. ML is basically a fitting procedure of pre-existing (experimental) data as a function of various factors called descriptors. If excellent descriptors can be selected and the training data contain negligible error, the predictive power of a ML model is relatively high. However, it is currently very difficult for a ML model to predict experimental results beyond the parameter space of the training experimental data. For example, it is pointed out that all-dislocation-ceramics, which could be a new type of solid electrolyte filled with appropriate dislocations for high ionic conductivity without dendrite formation, could not be predicted by ML. The merits and demerits of first principles calculations and physical or phenomenological ODE model calculations are also discussed with some examples of the flexoelectric effect, dielectric constant, and ionic conductivity in solid electrolytes.

Unraveling the Stability of Layered Intercalation Compounds through First-Principles Calculations: Establishing a Linear Free Energy Relationship with Aqueous Ions

Layered intercalation compounds, where atoms or molecules (intercalants) are inserted into layered materials (hosts), hold great potential for diverse applications. However, the lack of a systematic understanding of stable host-intercalant combinations poses challenges in materials design due to the vast combinatorial space. In this study, we performed first-principles calculations on 9024 compounds, unveiling a novel linear regression equation based on the principle of hard and soft acids and bases. This equation, incorporating the intercalant ion formation energy and ionic radius, quantitatively reveals the stability factors. Additionally, employing machine learning, we predicted regression coefficients from host properties, offering a comprehensive understanding and a predictive model for estimating the intercalation energy. Our work provides valuable insights into the energetics of layered intercalation compounds, facilitating targeted materials design.

Leveraging an all-fixed transfer framework to predict the interpretable formation energy of MXenes with hybrid terminals

MXenes have attracted substantial attention for their various applications in energy storage, sensors, and catalysts. Experimental exploration of MXenes with hybrid terminal surfaces offers a unique means of property tailoring that is crucial for expanding the performance space of MXenes, wherein the formation energy of an MXene with mixed surface terminals plays a key role in determining its relative stability and practical applications. However, the challenge of identifying energetically stable MXenes with multifunctional surfaces persists, primarily due to the absence of precise surface modification during experiments and the vast structural space for DFT calculations. Herein, we use an all-fixed transfer (AFT) framework combined with first-principles calculations to predict the formation energies of MXenes terminated by binary elements from groups VIA and VIIA. The trained model exhibits a high average R2 of 0.99, maintaining transferability and accuracy in predicting larger supercells from smaller-sized MXenes and datasets despite the structural imbalance between the training and prediction sets. The underlying interpretation of the high accuracy is revealed through the capture of main attributes and comparison of node features. Additionally, it is important to mention that the factors influencing the average formation energy include the types of element pairs, the ratio of terminal groups, and the distribution of terminations on two surfaces, with the first one being dominant. Finally, we successfully streamline the diverse structural cardinality of a large hybrid terminated MXene space of over 700 million, thereby facilitating the rapid screening of the top 5 stable MXene classes with binary terminal elements (FO, FCl, FBr, FS, and FSe). Besides, in the scenarios of lithium storage, the TL-predicted MXene can enhance its relative stability by increasing the fluorine ratio where the capacity can be optimized by different surface group combinations. All results indicate that the AFT framework has the advantages of screening functional MXenes with a huge structure space from smaller and imbalanced data sets.

The carbon footprint of predicting CO2 storage capacity in metal-organic frameworks within neural networks

While artificial intelligence drives remarkable progress in natural sciences, its broader societal implications are mostly disregarded. In this study, we evaluate environmental impacts of deep learning in materials science through extensive benchmarking. In particular, a set of diverse neural networks is trained for a given supervised learning task to assess greenhouse gas (GHG) emissions during training and inference phases. A chronological perspective showed diminishing returns, manifesting themselves as a 28% decrease in mean absolute error and nearly a 15,000% increase in the carbon footprint of model training in 2016-2022. By means of up-to-date graphics processing units, it is possible to partially offset the immense growth of GHG emissions. Nonetheless, the practice of employing energy-efficient hardware is overlooked by the materials informatics community, as follows from a literature analysis in the field. On the basis of our findings, we encourage researchers to report GHG emissions together with standard performance metrics.

Machine learning interatomic potential: Bridge the gap between small-scale models and realistic device-scale simulations

Machine learning interatomic potential (MLIP) overcomes the challenges of high computational costs in density-functional theory and the relatively low accuracy in classical large-scale molecular dynamics, facilitating more efficient and precise simulations in materials research and design. In this review, the current state of the four essential stages of MLIP is discussed, including data generation methods, material structure descriptors, six unique machine learning algorithms, and available software. Furthermore, the applications of MLIP in various fields are investigated, notably in phase-change memory materials, structure searching, material properties predicting, and the pre-trained universal models. Eventually, the future perspectives, consisting of standard datasets, transferability, generalization, and trade-off between accuracy and complexity in MLIPs, are reported.

Employing Graph Neural Networks for Predicting Electrode Average Voltages and Screening High-Voltage Sodium Cathode Materials

For many years, humans have been relentlessly focused on enhancing battery longevity and boosting energy storage capacities. The performance and durability of a battery depend significantly on the material used for its electrodes. In this context, merging machine learning with density functional theory (DFT) calculations has emerged as a pivotal approach to advancing the exploration of battery crystal structures. We present a new method that combines a graph convolutional neural network (GNN) with a Transformer convolutional layer, which we call Transformer-GNN. To underscore its efficacy, we benchmarked Transformer-GNN against three established statistical machine learning models: Support Vector Machine, Random Forest, and XGBoost. We also developed a standard GNN, which we refer to as Basic-GNN. Additionally, we compared Basic-GNN with Transformer-GNN to highlight the improvements brought about by incorporating the Transformer convolutional layer. The Transformer-GNN model outperforms the other models, achieving the highest R2 value of 0.82 and the lowest mean squared error of 0.3161. Our findings demonstrate that the Transformer-GNN can profoundly understand battery crystal structures, thus forging the path toward more sophisticated and durable battery systems. Leveraging the GNN model's voltage predictions in tandem with the capacity data sourced from the database, we screened and pinpointed Na(NiO2)2 as a high-voltage (higher than 5 V), high-capacity sodium cathode material. We conducted DFT calculations on Na(NiO2)2 and revealed the migration mechanism of the Na ions.

Accelerating material property prediction using generically complete isometry invariants

Periodic material or crystal property prediction using machine learning has grown popular in recent years as it provides a computationally efficient replacement for classical simulation methods. A crucial first step for any of these algorithms is the representation used for a periodic crystal. While similar objects like molecules and proteins have a finite number of atoms and their representation can be built based upon a finite point cloud interpretation, periodic crystals are unbounded in size, making their representation more challenging. In the present work, we adapt the Pointwise Distance Distribution (PDD), a continuous and generically complete isometry invariant for periodic point sets, as a representation for our learning algorithm. The PDD distinguished all (more than 660 thousand) periodic crystals in the Cambridge Structural Database as purely periodic sets of points without atomic types. We develop a transformer model with a modified self-attention mechanism that combines PDD with compositional information via a spatial encoding method. This model is tested on the crystals of the Materials Project and Jarvis-DFT databases and shown to produce accuracy on par with state-of-the-art methods while being several times faster in both training and prediction time.

The Future of Material Scientists in an Age of Artificial Intelligence

Material science has historically evolved in tandem with advancements in technologies for characterization, synthesis, and computation. Another type of technology to add to this mix is machine learning (ML) and artificial intelligence (AI). Now increasingly sophisticated AI-models are seen that can solve progressively harder problems across a variety of fields. From a material science perspective, it is indisputable that machine learning and artificial intelligence offer a potent toolkit with the potential to substantially accelerate research efforts in areas such as the development and discovery of new functional materials. Less clear is how to best harness this development, what new skill sets will be required, and how it may affect established research practices. In this paper, those question are explored with respect to increasingly more sophisticated ML/AI-approaches. To structure the discussion, a conceptual framework of an AI-ladder is introduced. This AI-ladder ranges from basic data-fitting techniques to more advanced functionalities such as semi-autonomous experimentation, experimental design, knowledge generation, hypothesis formulation, and the orchestration of specialized AI modules as stepping-stones toward general artificial intelligence. This ladder metaphor provides a hierarchical framework for contemplating the opportunities, challenges, and evolving skill sets required to stay competitive in the age of artificial intelligence.

Designing semiconductor materials and devices in the post-Moore era by tackling computational challenges with data-driven strategies

In the post-Moore's law era, the progress of electronics relies on discovering superior semiconductor materials and optimizing device fabrication. Computational methods, augmented by emerging data-driven strategies, offer a promising alternative to the traditional trial-and-error approach. In this Perspective, we highlight data-driven computational frameworks for enhancing semiconductor discovery and device development by elaborating on their advances in exploring the materials design space, predicting semiconductor properties and optimizing device fabrication, with a concluding discussion on the challenges and opportunities in these areas.

Machine learning in nanozymes: from design to application

Nanozymes, a distinctive class of nanomaterials endowed with enzyme-like activity and kinetics akin to enzyme-catalysed reactions, present several advantages over natural enzymes, including cost-effectiveness, heightened stability, and adjustable activity. However, the conventional trial-and-error methodology for developing novel nanozymes encounters growing challenges as research progresses. The advent of artificial intelligence (AI), particularly machine learning (ML), has ushered in innovative design approaches for researchers in this domain. This review delves into the burgeoning role of ML in nanozyme research, elucidating the advancements achieved through ML applications. The review explores successful instances of ML in nanozyme design and implementation, providing a comprehensive overview of the evolving landscape. A roadmap for ML-assisted nanozyme research is outlined, offering a universal guideline for research in this field. In the end, the review concludes with an analysis of challenges encountered and anticipates future directions for ML in nanozyme research. The synthesis of knowledge in this review aims to foster a cross-disciplinary study, propelling the revolutionary field forward.

Collaboration on Machine-Learned Potentials with IPSuite: A Modular Framework for Learning-on-the-Fly

The field of machine learning potentials has experienced a rapid surge in progress, thanks to advances in machine learning theory, algorithms, and hardware capabilities. While the underlying methods are continuously evolving, the infrastructure for their deployment has lagged. The community, due to these rapid developments, frequently finds itself split into groups built around different implementations of machine-learned potentials. In this work, we introduce IPSuite, a Python-driven software package designed to connect different methods and algorithms from the comprehensive field of machine-learned potentials into a single platform while also providing a collaborative infrastructure, helping ensure reproducibility. Furthermore, the data management infrastructure of the IPSuite code enables simple model sharing and deployment in simulations. Currently, IPSuite supports six state-of-the-art machine learning approaches for the fitting of interatomic potentials as well as a variety of methods for the selection of training data, running of ab initio calculations, learning-on-the-fly strategies, model evaluation, and simulation deployment.

Two-Dimensional Os2Se3 Nanosheet: A Ferroelectric Metal with Room-Temperature Ferromagnetism

Two-dimensional (2D) ferroelectric metals (FEMs) possess intriguing characteristics, such as unconventional superconductivity and the nonlinear anomalous Hall effect. However, their occurrence is exceedingly rare due to mutual repulsion between ferroelectricity and metallicity. In addition, further incorporating other features like ferromagnetism into FEMs to enhance their functionalities poses a significantly greater challenge. Here, via first-principles calculations, we demonstrate a case of an FEM that features a coexistence of room-temperature ferromagnetism, ferroelectricity, and metallicity in a thermodynamically stable 2D Os2Se3. It presents a vertical electric polarization of 3.00 pC/m that exceeds those of most FEMs and a moderate polarization switching barrier of 0.22 eV per formula unit. Moreover, 2D Os2Se3 exhibits robust ferromagnetism (Curie temperature TC ≈ 527 K) and a sizable magnetic anisotropy energy (-30.87 meV per formula unit). Furthermore, highly magnetization-dependent electrical conductivity is revealed, indicative of strong magnetoelectric coupling. Berry curvature calculation suggests that the FEM might exhibit nontrivial band topology.

Significant Impact of Defect Fluctuation on Charge Dynamics in CsPbI3: A Study Combining Machine Learning with Quantum Dynamics

In this study, we developed a machine-learned force field for CsPbI3 using a neural network potential, enabling molecular dynamics simulations (MD) with ab initio accuracy over nanoseconds. This approach, combined with ab initio MD and nonadiabatic MD, was used to study the charge trapping and recombination dynamics in both pristine and defective CsPbI3. Our simulations revealed key transitions affecting carrier lifetimes, especially in systems with iodine vacancy and interstitial iodine defects. An iodine trimer, formed when iodine replaces cesium, exhibits a high-frequency phonon mode. This mode enhances nonadiabatic coupling, accelerating charge recombination in defective systems compared to pristine ones. In the iodine vacancy system, recombination times varied significantly due to differences in NA coupling and energy gaps. The interplay between nonadiabatic coupling and pure dephasing time is crucial in determining recombination times for interstitial iodine defects. Our findings highlight the role of defect evolution in perovskites, offering insights for enhancing perovskite performance.

Band Alignment of Oxides by Learnable Structural-Descriptor-Aided Neural Network and Transfer Learning

The band alignment of semiconductors, insulators, and dielectrics is relevant to diverse material properties and device structures utilizing their surfaces and interfaces. In particular, the ionization potential and electron affinity are fundamental quantities that describe surface-dependent band-edge positions with respect to the vacuum level. Their accurate and systematic determination, however, demands elaborate experiments or simulations for well-characterized surfaces. Here, we report machine learning for the band alignment of nonmetallic oxides using a high-throughput first-principles calculation data set containing about 3000 oxide surfaces. Our neural network accurately predicts the band positions for relaxed surfaces of binary oxides simply by using the information on bulk structures and surface termination planes. Moreover, we extend the model to naturally include multiple-cation effects and transfer it to ternary oxides. The present approach enables the band alignment of a vast number of solid surfaces, thereby opening the way to a systematic understanding and materials screening.

Polymer-Unit Graph: Advancing Interpretability in Graph Neural Network Machine Learning for Organic Polymer Semiconductor Materials

The graph representation of complex materials plays a crucial role in the field of inorganic and organic materials investigations for developing data-centric materials science, such as those using graph neural networks (GNNs). However, the currently prevalent GNN models are primarily employed for investigating periodic crystals and organic small molecule data, yet they still encounter challenges in terms of interpretability and computational efficiency when applied to polymer monomers and organic macromolecules data. There is still a lack of graph representation of organic polymers and macromolecules specifically tailored for GNN models to explore the structural characteristics. The Polymer-unit Graph, a novel coarse-grained graph representation method introduced in study, is dedicated to expressing and analyzing polymers and macromolecules. By incorporating the Polymer-unit Graph into the GNN models and analyzing the organic semiconductor (OSC) materials database, it becomes possible to uncover intricate structure-property relationships involving branched-chain engineering, fluoridation substitution, and donor-acceptor combination effects on the elementary structure of OSC polymers. Furthermore, the Polymer-unit Graph enables visualizing the relationship between target properties and polymer units while reducing training time by an impressive 98% and minimizing molecular graph representation models. In conclusion, the Polymer-unit Graph successfully integrates the concept of Polymer-unit into the field of GNNs, enabling more accurate analysis and understanding of organic polymers and macromolecules.

Work Function Prediction by Graph Neural Networks for Configurationally Hybridized Boron-Doped Graphene

Graphene, serving as electrodes, is widely applied in electronic and optoelectronic devices. Work function as one of the fundamental intrinsic characteristics of graphene directly affects the interfacial properties of the electrodes, thereby affecting the performance of the devices. Much work has been done to regulate the work function of graphene to expand its application fields, and doping has been demonstrated as an effective method. However, the numerous types of doped graphene make the investigation of its work function time-consuming and labor-intensive. In order to quickly obtain the relationship between the structure and property, a deep learning method is employed to predict the work function in this study. Specifically, a data set of over 30,000 compositions with the work function on boron-doped graphene at different concentrations and doping positions via density functional theory simulations was established through ab initio calculations. Then, a novel fusion model (GT-Net) combining transformers and graph neural networks (GNNs) was proposed. After that, improved effective GNN-based descriptors were developed. Finally, three different GNN methods were compared, and the results show that the proposed method could accurately predicate the work function with the R2 = 0.975 and RMSE = 0.027. This study not only provides the possibility of designing materials with specific properties at the atomic level but also demonstrates the performance of GNNs on graph-level tasks with the same graph structure and atomic number.

When Machine Learning Meets 2D Materials: A Review

The availability of an ever-expanding portfolio of 2D materials with rich internal degrees of freedom (spin, excitonic, valley, sublattice, and layer pseudospin) together with the unique ability to tailor heterostructures made layer by layer in a precisely chosen stacking sequence and relative crystallographic alignments, offers an unprecedented platform for realizing materials by design. However, the breadth of multi-dimensional parameter space and massive data sets involved is emblematic of complex, resource-intensive experimentation, which not only challenges the current state of the art but also renders exhaustive sampling untenable. To this end, machine learning, a very powerful data-driven approach and subset of artificial intelligence, is a potential game-changer, enabling a cheaper - yet more efficient - alternative to traditional computational strategies. It is also a new paradigm for autonomous experimentation for accelerated discovery and machine-assisted design of functional 2D materials and heterostructures. Here, the study reviews the recent progress and challenges of such endeavors, and highlight various emerging opportunities in this frontier research area.

Optical property dataset of inorganic phosphor

Developing inorganic phosphor with desired properties for light-emitting diode application has traditionally relied on time-consuming and labor-intensive material development processes. Moreover, the results of material development research depend significantly on individual researchers' intuition and experience. Thus, to improve the efficiency and reliability of materials discovery, machine learning has been widely applied to various materials science applications in recent years. However, the prediction capabilities of machine learning methods fundamentally depend on the quality of the training datasets. In this work, we constructed a high-quality and reliable dataset that contains experimentally validated inorganic phosphors and their optical properties, sourced from the literature on inorganic phosphors. Our dataset includes 3952 combinations of 21 dopant elements in 2238 host materials from 553 articles. The dataset provides material information, optical properties, measurement conditions for inorganic phosphors, and meta-information. Among the preliminary machine learning results, the essential properties of inorganic phosphors, such as maximum Photoluminescence (PL) emission wavelength and PL decay time, show overall satisfactory prediction performance with coefficient of determination ( R 2 ) scores of 0.7 or more. We also confirmed that the measurement conditions significantly improved prediction performance.

Machine-Learning-Assisted Construction of Ternary Convex Hull Diagrams

In the search for novel intermetallic ternary alloys, much of the effort goes into performing a large number of ab initio calculations covering a wide range of compositions and structures. These are essential to building a reliable convex hull diagram. While density functional theory (DFT) provides accurate predictions for many systems, its computational overheads set a throughput limit on the number of hypothetical phases that can be probed. Here, we demonstrate how an ensemble of machine-learning (ML) spectral neighbor-analysis potentials (SNAPs) can be integrated into a workflow for the construction of accurate ternary convex hull diagrams, highlighting regions that are fertile for materials discovery. Our workflow relies on using available binary-alloy data both to train the SNAP models and to create prototypes for ternary phases. From the prototype structures, all unique ternary decorations are created and used to form a pool of candidate compounds. The SNAPs ensemble is then used to prerelax the structures and screen the most favorable prototypes before using DFT to build the final phase diagram. As constructed, the proposed workflow relies on no extra first-principles data to train the ML surrogate model and yields a DFT-level accurate convex hull. We demonstrate its efficacy by investigating the Cu-Ag-Au and Mo-Ta-W ternary systems.

Coarse-Grained Crystal Graph Neural Networks for Reticular Materials Design

Reticular materials, including metal-organic frameworks and covalent organic frameworks, combine the relative ease of synthesis and an impressive range of applications in various fields from gas storage to biomedicine. Diverse properties arise from the variation of building units─metal centers and organic linkers─in almost infinite chemical space. Such variation substantially complicates the experimental design and promotes the use of computational methods. In particular, the most successful artificial intelligence algorithms for predicting the properties of reticular materials are atomic-level graph neural networks, which optionally incorporate domain knowledge. Nonetheless, the data-driven inverse design involving these models suffers from the incorporation of irrelevant and redundant features such as a full atomistic graph and network topology. In this study, we propose a new way of representing materials, aiming to overcome the limitations of existing methods; the message passing is performed on a coarse-grained crystal graph that comprises molecular building units. To highlight the merits of our approach, we assessed the predictive performance and energy efficiency of neural networks built on different materials representations, including composition-based and crystal-structure-aware models. Coarse-grained crystal graph neural networks showed decent accuracy at low computational costs, making them a valuable alternative to omnipresent atomic-level algorithms. Moreover, the presented models can be successfully integrated into an inverse materials design pipeline as estimators of the objective function. Overall, the coarse-grained crystal graph framework is aimed at challenging the prevailing atom-centric perspective on reticular materials design.

Electronic Properties of CrB/Co2CO2 Superlattices by Multiple Descriptor-Based Machine Learning Combined with First-Principles

In recent times, newly unveiled 2D materials exhibiting exceptional characteristics, such as MBenes and MXenes, have gained widespread application across diverse domains, encompassing electronic devices, catalysis, energy storage, sensors, and various others. Nonetheless, numerous technical bottlenecks persist in the development of high-performance, structurally flexible, and adjustable electronic device materials. Research investigations have demonstrated that 2D van der Waals superlattices (vdW SLs) structures comprising materials exhibit exceptional electrical, mechanical, and optical properties. In this work, the advantages of both materials are combined and compose the vdW SLs structure of MBenes and MXenes, thus obtaining materials with excellent electronic properties. Furthermore, it integrates machine learning (ML) with first-principles methods to forecast the electrical properties of MBene/MXene superlattice materials. Initially, various configurations of MBene/MXene superlattice materials are explored, revealing that distinct stacking methods exert significant influence on the electronic structure of MBene/MXene materials. Specifically, the BABA-type stacking of CrB (layer A) and Co2CO2 MXene (layer B) is most stable configureation. Subsequently, multiple descriptors of the structure are constructed to predict the density of states  of vdW SLs through the employment of ML techniques. The best model achieves a mean absolute error (MAE) as low as 0.147 eV.

Enhanced Sampling of Crystal Nucleation with Graph Representation Learnt Variables

In this study, we present a graph neural network (GNN)-based learning approach using an autoencoder setup to derive low-dimensional variables from features observed in experimental crystal structures. These variables are then biased in enhanced sampling to observe state-to-state transitions and reliable thermodynamic weights. In our approach, we used simple convolution and pooling methods. To verify the effectiveness of our protocol, we examined the nucleation of various allotropes and polymorphs of iron and glycine in their molten states. Our graph latent variables, when biased in well-tempered metadynamics, consistently show transitions between states and achieve accurate thermodynamic rankings in agreement with experiments, both of which are indicators of dependable sampling. This underscores the strength and promise of our GNN variables for improved sampling. The protocol shown here should be applicable for other systems and other sampling methods.

Machine Learning in Membrane Design: From Property Prediction to AI-Guided Optimization

Porous membranes, either polymeric or two-dimensional materials, have been extensively studied because of their outstanding performance in many applications such as water filtration. Recently, inspired by the significant success of machine learning (ML) in many areas of scientific discovery, researchers have started to tackle the problem in the field of membrane design using data-driven ML tools. In this Mini Review, we summarize research efforts on three types of applications of machine learning in membrane design, including (1) membrane property prediction using ML, (2) gaining physical insight and drawing quantitative relationships between membrane properties and performance using explainable artificial intelligence, and (3) ML-guided design, optimization, or virtual screening of membranes. On top of the review of previous research, we discuss the challenges associated with applying ML for membrane design and potential future directions.

Charge-Optimized Electrostatic Interaction Atom-Centered Neural Network Algorithm

Machine-learning algorithms have been proposed to capture electrostatic interactions by using effective partial charges. These algorithms often rely on a pretrained model for partial charge prediction using density functional theory-calculated partial charges as references, which introduces complexity to the force field model. The accuracy of the trained model also depends on the reliability of charge partition methods, which can be dependent on the specific system and methodology employed. In this study, we propose an atom-centered neural network (ANN) algorithm that eliminates the need for reference charges. Our algorithm requires only a single NN model for each element to obtain both atomic energy and charges. These atomic charges are then employed to compute electrostatic energies using the Ewald summation algorithm. Subsequently, the force field model is trained on total energy and forces, with the inclusion of electrostatic energy. To evaluate the performance of our algorithm, we conducted tests on three benchmark systems, including a Ge slab with an O adatom system, a TiO2 crystalline system, and a Pd-O nanoparticle system. Our results demonstrate reasonably accurate predictions of partial charges and electrostatic interactions. This algorithm provides a self-consistent charge prediction strategy and possibilities for robust and reliable modeling of electrostatic interactions in machine-learning potentials.

Automatic Prediction of Peak Optical Absorption Wavelengths in Molecules Using Convolutional Neural Networks

Molecular design depends heavily on optical properties for applications such as solar cells and polymer-based batteries. Accurate prediction of these properties is essential, and multiple predictive methods exist, from ab initio to data-driven techniques. Although theoretical methods, such as time-dependent density functional theory (TD-DFT) calculations, have well-established physical relevance and are among the most popular methods in computational physics and chemistry, they exhibit errors that are inherent in their approximate nature. These high-throughput electronic structure calculations also incur a substantial computational cost. With the emergence of big-data initiatives, cost-effective, data-driven methods have gained traction, although their usability is highly contingent on the degree of data quality and sparsity. In this study, we present a workflow that employs deep residual convolutional neural networks (DR-CNN) and gradient boosting feature selection to predict peak optical absorption wavelengths (λmax) exclusively from SMILES representations of dye molecules and solvents; one would normally measure λmax using UV-vis absorption spectroscopy. We use a multifidelity modeling approach, integrating 34,893 DFT calculations and 26,395 experimentally derived λmax data, to deliver more accurate predictions via a Bayesian-optimized gradient boosting machine. Our approach is benchmarked against the state of the art that is reported in the scientific literature; results demonstrate that learnt representations via a DR-CNN workflow that is integrated with other machine learning methods can accelerate the design of molecules for specific optical characteristics.

Machine-learning accelerated structure search for ligand-protected clusters

Finding low-energy structures of ligand-protected clusters is challenging due to the enormous conformational space and the high computational cost of accurate quantum chemical methods for determining the structures and energies of conformers. Here, we adopted and utilized a kernel rigid regression based machine learning method to accelerate the search for low-energy structures of ligand-protected clusters. We chose the Au25(Cys)18 (Cys: cysteine) cluster as a model system to test and demonstrate our method. We found that the low-energy structures of the cluster are characterized by a specific hydrogen bond type in the cysteine. The different configurations of the ligand layer influence the structural and electronic properties of clusters.

Informative Training Data for Efficient Property Prediction in Metal-Organic Frameworks by Active Learning

In recent data-driven approaches to material discovery, scenarios where target quantities are expensive to compute and measure are often overlooked. In such cases, it becomes imperative to construct a training set that includes the most diverse, representative, and informative samples. Here, a novel regression tree-based active learning algorithm is employed for such a purpose. It is applied to predict the band gap and adsorption properties of metal-organic frameworks (MOFs), a novel class of materials that results from the virtually infinite combinations of their building units. Simpler and low dimensional descriptors, such as those based on stoichiometric and geometric properties, are used to compute the feature space for this model owing to their ability to better represent MOFs in the low data regime. The partitions given by a regression tree constructed on the labeled part of the data set are used to select new samples to be added to the training set, thereby limiting its size while maximizing the prediction quality. Tests on the QMOF, hMOF, and dMOF data sets reveal that our method constructs small training data sets to learn regression models that predict the target properties more efficiently than existing active learning approaches, and with lower variance. Specifically, our active learning approach is highly beneficial when labels are unevenly distributed in the descriptor space and when the label distribution is imbalanced, which is often the case for real world data. The regions defined by the tree help in revealing patterns in the data, thereby offering a unique tool to efficiently analyze complex structure-property relationships in materials and accelerate materials discovery.

Quest for Multifunctionality: Current Progress in the Characterization of Heterometallic Metal-Organic Frameworks

Metal-organic frameworks (MOFs) are a class of porous, crystalline materials that have been systematically developed for a broad range of applications. Incorporation of two or more metals into a single crystalline phase to generate heterometallic MOFs has been shown to lead to synergistic effects, in which the whole is oftentimes greater than the sum of its parts. Because geometric proximity is typically required for metals to function cooperatively, deciphering and controlling metal distributions in heterometallic MOFs is crucial to establish structure-function relationships. However, determination of short- and long-range metal distributions is nontrivial and requires the use of specialized characterization techniques. Advancements in the characterization of metal distributions and interactions at these length scales is key to rapid advancement and rational design of functional heterometallic MOFs. This perspective summarizes the state-of-the-art in the characterization of heterometallic MOFs, with a focus on techniques that allow metal distributions to be better understood. Using complementary analyses, in conjunction with computational methods, is critical as this field moves toward increasingly complex, multifunctional systems.

Physics-Guided Dual Self-Supervised Learning for Structure-Based Material Property Prediction

Deep learning models have been widely used for high-performance material property prediction. However, training such models usually requires a large amount of labeled data, which are usually unavailable. Self-supervised learning (SSL) methods have been proposed to address this data scarcity issue. Herein, we present DSSL, a physics-guided dual SSL framework, for graph neural network-based material property prediction, which combines node masking-based generative SSL with atomic coordinate perturbation-based contrastive SSL strategies to capture local and global information about input crystals. Moreover, we achieve physics-guided pretraining by using the macroproperty (e.g., elasticity)-related microproperty prediction of atomic stiffness as an additional pretext task. We pretrain our DSSL model on the Materials Project database and fine-tune it with 10 material property data sets. The experimental results demonstrate that teaching neural networks some physics using the SSL strategy can afford ≤26.89% performance improvement compared to that of the baseline models.

PNAbind: Structure-based prediction of protein-nucleic acid binding using graph neural networks

The recognition and binding of nucleic acids (NAs) by proteins depends upon complementary chemical, electrostatic and geometric properties of the protein-NA binding interface. Structural models of protein-NA complexes provide insights into these properties but are scarce relative to models of unbound proteins. We present a deep learning approach for predicting protein-NA binding given the apo structure of a protein (PNAbind). Our method utilizes graph neural networks to encode spatial distributions of physicochemical and geometric properties of the protein molecular surface that are predictive of NA binding. Using global physicochemical encodings, our models predict the overall binding function of a protein and can discriminate between specificity for DNA or RNA binding. We show that such predictions made on protein structures modeled with AlphaFold2 can be used to gain mechanistic understanding of chemical and structural features that determine NA recognition. Using local encodings, our models predict the location of NA binding sites at the level of individual binding residues. Binding site predictions were validated against benchmark datasets, achieving AUROC scores in the range of 0.92-0.95. We applied our models to the HIV-1 restriction factor APOBEC3G and show that our predictions are consistent with experimental RNA binding data.

The Case for a Defect Genome Initiative

The Materials Genome Initiative (MGI) has streamlined the materials discovery effort by leveraging generic traits of materials, with focus largely on perfect solids. Defects such as impurities and perturbations, however, drive many attractive functional properties of materials. The rich tapestry of charge, spin, and bonding states hosted by defects are not accessible to elements and perfect crystals, and defects can thus be viewed as another class of "elements" that lie beyond the periodic table. Accordingly, a Defect Genome Initiative (DGI) to accelerate functional defect discovery for energy, quantum information, and other applications is proposed. First, major advances made under the MGI are highlighted, followed by a delineation of pathways for accelerating the discovery and design of functional defects under the DGI. Near-term goals for the DGI are suggested. The construction of open defect platforms and design of data-driven functional defects, along with approaches for fabrication and characterization of defects, are discussed. The associated challenges and opportunities are considered and recent advances towards controlled introduction of functional defects at the atomic scale are reviewed. It is hoped this perspective will spur a community-wide interest in undertaking a DGI effort in recognition of the importance of defects in enabling unique functionalities in materials.

A comprehensive transformer-based approach for high-accuracy gas adsorption predictions in metal-organic frameworks

Gas separation is crucial for industrial production and environmental protection, with metal-organic frameworks (MOFs) offering a promising solution due to their tunable structural properties and chemical compositions. Traditional simulation approaches, such as molecular dynamics, are complex and computationally demanding. Although feature engineering-based machine learning methods perform better, they are susceptible to overfitting because of limited labeled data. Furthermore, these methods are typically designed for single tasks, such as predicting gas adsorption capacity under specific conditions, which restricts the utilization of comprehensive datasets including all adsorption capacities. To address these challenges, we propose Uni-MOF, an innovative framework for large-scale, three-dimensional MOF representation learning, designed for multi-purpose gas prediction. Specifically, Uni-MOF serves as a versatile gas adsorption estimator for MOF materials, employing pure three-dimensional representations learned from over 631,000 collected MOF and COF structures. Our experimental results show that Uni-MOF can automatically extract structural representations and predict adsorption capacities under various operating conditions using a single model. For simulated data, Uni-MOF exhibits remarkably high predictive accuracy across all datasets. Additionally, the values predicted by Uni-MOF correspond with the outcomes of adsorption experiments. Furthermore, Uni-MOF demonstrates considerable potential for broad applicability in predicting a wide array of other properties.

Deep-Learning Density Functional Perturbation Theory

Calculating perturbation response properties of materials from first principles provides a vital link between theory and experiment, but is bottlenecked by the high computational cost. Here, a general framework is proposed to perform density functional perturbation theory (DFPT) calculations by neural networks, greatly improving the computational efficiency. Automatic differentiation is applied on neural networks, facilitating accurate computation of derivatives. High efficiency and good accuracy of the approach are demonstrated by studying electron-phonon coupling and related physical quantities. This work brings deep-learning density functional theory and DFPT into a unified framework, creating opportunities for developing ab initio artificial intelligence.

Automatic Prediction of Band Gaps of Inorganic Materials Using a Gradient Boosted and Statistical Feature Selection Workflow

Machine learning (ML) methods can train a model to predict material properties by exploiting patterns in materials databases that arise from structure-property relationships. However, the importance of ML-based feature analysis and selection is often neglected when creating such models. Such analysis and selection are especially important when dealing with multifidelity data because they afford a complex feature space. This work shows how a gradient-boosted statistical feature-selection workflow can be used to train predictive models that classify materials by their metallicity and predict their band gap against experimental measurements, as well as computational data that are derived from electronic-structure calculations. These models are fine-tuned via Bayesian optimization, using solely the features that are derived from chemical compositions of the materials data. We test these models against experimental, computational, and a combination of experimental and computational data. We find that the multifidelity modeling option can reduce the number of features required to train a model. The performance of our workflow is benchmarked against state-of-the-art algorithms, the results of which demonstrate that our approach is either comparable to or superior to them. The classification model realized an accuracy score of 0.943, a macro-averaged F1-score of 0.940, area under the curve of the receiver operating characteristic curve of 0.985, and an average precision of 0.977, while the regression model achieved a mean absolute error of 0.246, a root-mean squared error of 0.402, and R2 of 0.937. This illustrates the efficacy of our modeling approach and highlights the importance of thorough feature analysis and judicious selection over a "black-box" approach to feature engineering in ML-based modeling.

Graph attention neural networks for mapping materials and molecules beyond short-range interatomic correlations

Bringing advances in machine learning to chemical science is leading to a revolutionary change in the way of accelerating materials discovery and atomic-scale simulations. Currently, most successful machine learning schemes can be largely traced to the use of localized atomic environments in the structural representation of materials and molecules. However, this may undermine the reliability of machine learning models for mapping complex systems and describing long-range physical effects because of the lack of non-local correlations between atoms. To overcome such limitations, here we report a graph attention neural network as a unified framework to map materials and molecules into a generalizable and interpretable representation that combines local and non-local information of atomic environments from multiple scales. As an exemplary study, our model is applied to predict the electronic structure properties of metal-organic frameworks (MOFs) which have notable diversity in compositions and structures. The results show that our model achieves the state-of-the-art performance. The clustering analysis further demonstrates that our model enables high-level identification of MOFs with spatial and chemical resolution, which would facilitate the rational design of promising reticular materials. Furthermore, the application of our model in predicting the heat capacity of complex nanoporous materials, a critical property in a carbon capture process, showcases its versatility and accuracy in handling diverse physical properties beyond electronic structures.