Machine-Learning-Driven High-Throughput Screening for High-Energy Density and Stable NASICON Cathodes.

The Na super ionic conductor (NASICON), which has outstanding structural stability and a high operating voltage, is an appealing material for overcoming the limits of low specific energy and larger volume distortion of sodium-ion batteries. In this study, to discover ideal NASICON cathode materials, a screening platform based on density functional theory (DFT) calculations and machine learning (ML) is developed. A training database was generated utilizing the previous 124 545 electrode databases, and a test set of 3126 potential NASICON structures [NaxMyM'1-y(PO4)3] with 27 dopants at the metal site and 6 dopants at the polyanion central site was constructed. The developed ML surrogate model identifies 796 materials that satisfy the following criteria: formation energy of <0.0 eV/atom, energy above hull of ≤0.025 eV/atom, volume change of ≤4%, and theoretical capacity of ≥50 mAh/g. The thermodynamically stable configurations of doped NASICON structures were then selected using machine learning interatomic potential (MLIP), enabling rapid consideration of various dopant site configurations. DFT calculations are followed on 796 screened materials to obtain energy density, average voltage, and volume change. Finally, 50 candidates with an average voltage of ≥3.5 V are identified. The suggested platform accelerates the exploration for optimal NASICON materials by narrowing the focus on materials with desired properties, saving considerable resources.

Exploring the large chemical space in search of thermodynamically stable and mechanically robust MXenes via machine learning.

To effectively utilize MXenes, a family of two-dimensional materials, in various applications that include thermoelectric devices, semiconductors, and transistors, their thermodynamic and mechanical properties, which are closely related to their stability, must be understood. However, exploring the large chemical space of MXenes and verifying their stability using first-principles calculations are computationally expensive and inefficient. Therefore, this study proposes a machine learning (ML)-based high-throughput MXene screening framework to identify thermodynamically stable MXenes and determine their mechanical properties. A dataset of 23 857 MXenes with various compositions was used to validate this framework, and 48 MXenes were predicted to be stable by ML models in terms of heat of formation and energy above the convex hull. Among them, 45 MXenes were validated using density functional theory calculations, of which 23 MXenes, including Ti2CClBr and Zr2NCl2, have not been previously known for their stability, confirming the effectiveness of this framework. The in-plane stiffness, shear moduli, and Poisson's ratio of the 45 MXenes were observed to vary widely according to their constituent elements, ranging from 90.11 to 198.02 N m-1, 64.00 to 163.40 N m-1, and 0.19 to 0.58, respectively. MXenes with Group-4 transition metals and halogen surface terminations were shown to be both thermodynamically stable and mechanically robust, highlighting the importance of electronegativity difference between constituent elements. Structurally, a smaller volume per atom and minimum bond length were determined to be preferable for obtaining mechanically robust MXenes. The proposed framework, along with an analysis of these two properties of MXenes, demonstrates immense potential for expediting the discovery of stable and robust MXenes.

Atomistic insights into adhesion characteristics of tungsten on titanium nitride using steered molecular dynamics with machine learning interatomic potential.

As transistor integration accelerates and miniaturization progresses, improving the interfacial adhesion characteristics of complex metal interconnect has become a major issue in ensuring semiconductor device reliability. Therefore, it is becoming increasingly important to interpret the adhesive properties of metal interconnects at the atomic level, predict their adhesive strength and failure mode, and develop computational methods that can be universally applied regardless of interface properties. In this study, we propose a method for theoretically understanding adhesion characteristics through steering molecular dynamics simulations based on machine learning interatomic potentials. We utilized this method to investigate the adhesion characteristics of tungsten deposited on titanium nitride barrier metal (W/TiN) as a representative metal interconnect structure in devices. Pulling tests that pull two materials apart and sliding tests that pull them against each other in a shear direction were implemented to investigate the failure mode and adhesive strength depending on TiN facet orientation. We found that the W/TiN interface showed an adhesive failure where they separate from each other when tested with pulling force on Ti-rich (111) or (001) facets while cohesive failures occurred where W itself was destroyed on N-rich (111) facet. The adhesion strength was defined as the maximum force causing failure during the pulling test for consistent interpretation and the strengths of tungsten were predicted to be strongest when deposited onto N-rich (111) facet while weakest on Ti-rich (111) facet.

Prediction of Protein Aggregation Propensity via Data-Driven Approaches.

Protein aggregation occurs when misfolded or unfolded proteins physically bind together and can promote the development of various amyloid diseases. This study aimed to construct surrogate models for predicting protein aggregation via data-driven methods using two types of databases. First, an aggregation propensity score database was constructed by calculating the scores for protein structures in the Protein Data Bank using Aggrescan3D 2.0. Moreover, feature- and graph-based models for predicting protein aggregation have been developed by using this database. The graph-based model outperformed the feature-based model, resulting in an R2 of 0.95, although it intrinsically required protein structures. Second, for the experimental data, a feature-based model was built using the Curated Protein Aggregation Database 2.0 to predict the aggregated intensity curves. In summary, this study suggests approaches that are more effective in predicting protein aggregation, depending on the type of descriptor and the database.

AmorProt: Amino Acid Molecular Fingerprints Repurposing-Based Protein Fingerprint.

As protein therapeutics play an important role in almost all medical fields, numerous studies have been conducted on proteins using artificial intelligence. Artificial intelligence has enabled data-driven predictions without the need for expensive experiments. Nevertheless, unlike the various molecular fingerprint algorithms that have been developed, protein fingerprint algorithms have rarely been studied. In this study, we proposed the amino acid molecular fingerprints repurposing-based protein (AmorProt) fingerprint, a protein sequence representation method that effectively uses the molecular fingerprints corresponding to 20 amino acids. Subsequently, the performances of the tree-based machine learning and artificial neural network models were compared using (1) amyloid classification and (2) isoelectric point regression. Finally, the applicability and advantages of the developed platform were demonstrated through a case study and the following experiments: (3) comparison of dataset dependence with feature-based methods, (4) feature importance analysis, and (5) protein space analysis. Consequently, the significantly improved model performance and data-set-independent versatility of the AmorProt fingerprint were verified. The results revealed that the current protein representation method can be applied to various fields related to proteins, such as predicting their fundamental properties or interaction with ligands.

Screening Platform for Promising Na Superionic Conductors for Na-Ion Solid-State Electrolytes.

Na-ion batteries are considered a promising alternative to the analogous Li-ion batteries because of their low manufacturing cost, large abundance, and similar chemical/electrochemical properties. In particular, research on Na-ion solid electrolytes, which resolve the flammability issues associated with liquid electrolytes and increase the energy density obtained using a particular metal anode, is rapidly growing. However, the ionic conductivities of these materials are lower than those of liquids. We present a novel classification approach based on machine learning for identifying Na superionic conductor (NASICON) materials with outstanding ionic conductivities. We obtained new features based on chemical descriptors such as Na content, elemental radii, and electronegativity. We then classified 3573 NASICON structures by implementing the ensemble model of gradient boosting algorithms, with an average prediction accuracy of 84.2%. We further validated the thermodynamic stability and ionic conductivity values of the materials classified as superionic materials by employing density functional theory calculations and ab initio molecular dynamics simulations. Na3YTaSi2PO12, Na3HfZrSi2PO12, Na3LaTaSi2PO12, and Na3ScTaSi2PO12 were confirmed as promising NASICON structures that fulfill the requirements of solid-state electrolytes.

Enhanced Hydrogen Evolution Performance at the Lateral Interface between Two Layered Materials Predicted with Machine Learning.

While economical and effective catalysts are required for sustainable hydrogen production, low-dimensional interfacial engineering techniques have been developed to improve the catalytic activity in the hydrogen evolution reaction (HER). In this study, we used density functional theory (DFT) calculations to measure the Gibbs free energy change (ΔGH) in hydrogen adsorption in two-dimensional lateral heterostructures (LHSs) MX2/M'X'2 (MoS2/WS2, MoS2/WSe2, MoSe2/WS2, MoSe2/WSe2, MoTe2/WSe2, MoTe2/WTe2, and WS2/WSe2) and MX2/M'X' (NbS2/ZnO, NbSe2/ZnO, NbS2/GaN, MoS2/ZnO, MoSe2/ZnO, MoS2/AlN, MoS2/GaN, and MoSe2/GaN) at several different positions near the interface. Compared to the interfaces of LHS MX2/M'X'2 and the surfaces of the monolayer MX2 and MX, the interfaces of LHS MX2/M'X' display greater hydrogen evolution reactivity due to their metallic behavior. The hydrogen absorption is stronger at the interfaces of LHS MX2/M'X', and that facilitates proton accessibility and increases the usage of catalytically active sites. Here, we develop three types of descriptors that can be used universally in 2D materials and can explain changes in ΔGH for different adsorption sites in a single LHS using only the basic information of the LHSs (type and number of neighboring atoms to the adsorption points). Using the DFT results of the LHSs and the various experimental data of atomic information, we trained machine learning (ML) models with the chosen descriptors to predict promising combinations and adsorption sites for HER catalysts among the LHSs. Our ML model achieved an R2 score of 0.951 (regression) and an F1 score of 0.749 (classification). Furthermore, the developed surrogate model was implemented to predict the structures in the test set and was based on confirmation from the DFT calculations via ΔGH values. The LHS MoS2/ZnO is the best candidate for HER among 49 candidates considered using both DFT and ML models because it has a ΔGH of -0.02 eV on top of O at the interface position and requires only -171 mV of overpotential to obtain the standard current density (10 A/cm2).

Extracting Chemical Information from Scientific Literature Using Text Mining: Building an Ionic Conductivity Database for Solid-State Electrolytes.

Recently, as the demand for electric vehicles has rapidly grown, concerns regarding the safety of liquid electrolytes used as battery materials have increased. Rechargeable batteries made of liquid electrolytes pose a risk of fire and may explode due to the decomposition reaction of the electrolyte. Accordingly, interest in solid-state electrolytes (SSEs), which have greater stability than liquid electrolytes, is increasing, and research into finding stable SSEs with high ionic conductivity is actively being conducted. Consequently, it is essential to obtain a large amount of material data to explore new SSEs. However, the data collection process is highly repetitive and time-consuming. Therefore, the goal of this study is to automatically extract the ionic conductivities of SSEs from published literature using text-mining algorithms and use this information to construct a materials database. The extraction procedure includes document processing, natural language preprocessing, phase parsing, relation extraction, and data post-processing. For performance verification, the ionic conductivities are extracted from 38 studies, and the accuracy of the proposed model is confirmed by comparing extracted conductivities with the actual ones. In previous research, 93% of battery-related records were unable to distinguish between ionic and electrical conductivities. However, by applying the proposed model, the proportion of undistinguished records was successfully reduced from 93 to 24.3%. Finally, the ionic conductivity database was constructed by extracting the ionic conductivity from 3258 papers, and the battery database was reconstructed by adding eight pieces of representative structural information.

Accelerated Discovery of Novel Garnet-Type Solid-State Electrolyte Candidates via Machine Learning.

All-solid-state batteries (ASSBs) have attracted considerable attention because of their higher energy density and stability than conventional lithium-ion batteries (LIBs). For the development of promising ASSBs, solid-state electrolytes (SSEs) are essential to achieve structural integrity. Thus, in this study, a machine-learning-based surrogate model was developed to search for ideal garnet-type SSE candidates. The well-known Li7La3Zr2O12 structure was used as a base material, and 73 chemical elements were substituted on La and Zr sites, leading to 5329 potential structures. First, the elasticity database and machine learning descriptors were adopted from previous studies. Subsequently, the machine-learning-based surrogate model was applied to predict the elastic properties of potential SSE materials, followed by first-principles calculations for validation. Furthermore, the active learning process demonstrated that it can effectively decrease prediction uncertainty. Finally, the ionic conductivity of the mechanically superior materials was predicted to suggest optimal SSE candidates. Then, ab initio molecular dynamics simulations are followed for confirmation of diffusion behavior for materials classified as superionic; 10 new tetragonal-phase garnet SSEs are verified with superior mechanical and ionic conductivity properties. We believe that the current model and the constructed database will become a cornerstone for the development of next-generation SSE materials.

Optimal machine learning feature selection for assessing the mechanical properties of a zeolite framework.

In this study, 45 and 249 critical features were discovered among 896 zeolite descriptors generated by the matminer package for estimating the shear and bulk moduli of zeolites, respectively. A database containing the mechanical properties of 873 zeolite structures, calculated using density functional theory, was used to train the machine learning regression model. The results of using these critical features with the LightGBM algorithm were rigorously compared with those from other regressors as well as with different sets of features. The comparison results indicate that the surrogate model with critical features increases the prediction accuracy of the bulk and shear moduli of zeolites by 17.3% and 10.6%, respectively, and reduces the prediction uncertainty by one-third of that achieved using previously available features. The suggested features originating from several physical and chemical groups highlight the unveiled relationships between the features and mechanical properties of zeolites. The robustness of the constructed model with 356 features was confirmed by applying a set of different training-test set ratios. We believe that the suggested critical features of zeolites can help to understand the mechanical behavior of a half million unlabeled hypothetical zeolite structures and accelerate the discovery of novel zeolites with unprecedented mechanical properties.

MGCVAE: Multi-Objective Inverse Design via Molecular Graph Conditional Variational Autoencoder.

The ultimate goal of various fields is to directly generate molecules with desired properties, such as water-soluble molecules in drug development and molecules suitable for organic light-emitting diodes or photosensitizers in the field of development of new organic materials. This study proposes a molecular graph generative model based on an autoencoder for the de novo design. The performance of the molecular graph conditional variational autoencoder (MGCVAE) for generating molecules with specific desired properties was investigated by comparing it to a molecular graph variational autoencoder (MGVAE). Furthermore, multi-objective optimization for MGCVAE was applied to satisfy the two selected properties simultaneously. In this study, two physical properties, calculated logP and molar refractivity, were used as optimization targets for designing de novo molecules. Consequently, it was confirmed that among the generated molecules, 25.89% of the optimized molecules were generated in MGCVAE compared to 0.66% in MGVAE. This demonstrates that MGCVAE effectively produced drug-like molecules with two target properties. The results of this study suggest that these graph-based data-driven models are an effective method for designing new molecules that fulfill various physical properties.

Dual doping with cations and anions for enhancing the structural stability of the sodium-ion layered cathode.

First-principles-based calculations were implemented to explore the ideal combination of cations and anions as dual dopants for enhancing the structural stability of the sodium-ion layered cathode for application in sodium ion batteries (SIBs), leading to improved electrochemical properties. Cation-doped NaNi0.42Mn0.5D0.08O2 was chosen as the reference structure, where D represents twelve cation dopants (Ga, Ge, Hf, In, Pt, Rh, Ru, Sb, Te, Ti, Y, and Zr), which have been proven to have excellent performance. Fluoride was selected as the anion dopant to give the general formula NaNi0.42Mn0.5D0.08O1.96F0.04, leading to a total of twelve different combinations of cation and anion co-doped structures. The screening criteria include the formation energy, which was used to confirm the thermodynamically favored locations of the dopants; the phase stability; and the volume change accompanying the transformation from the O3- to P3-phase after 50% desodiation. The calculations show that Te-, Sb-, Hf-, Y-, and Ti-F are the five most effective dual dopants for potentially enhancing the structural stability of the sodium-ion layered oxide during cycling. The present study provides an essential design map for developing an ideal dual doping strategy for SIBs.

Novel Solubility Prediction Models: Molecular Fingerprints and Physicochemical Features vs Graph Convolutional Neural Networks.

Predicting both accurate and reliable solubility values has long been a crucial but challenging task. In this work, surrogated model-based methods were developed to accurately predict the solubility of two molecules (solute and solvent) through machine learning and deep learning. The current study employed two methods: (1) converting molecules into molecular fingerprints and adding optimal physicochemical properties as descriptors and (2) using graph convolutional network (GCN) models to convert molecules into a graph representation and deal with prediction tasks. Then, two prediction tasks were conducted with each method: (1) the solubility value (regression) and (2) the solubility class (classification). The fingerprint-based method clearly demonstrates that high performance is possible by adding simple but significant physicochemical descriptors to molecular fingerprints, while the GCN method shows that it is possible to predict various properties of chemical compounds with relatively simplified features from the graph representation. The developed methodologies provide a comprehensive understanding of constructing a proper model for predicting solubility and can be employed to find suitable solutes and solvents.

Prediction of dielectric constants of ABO3-type perovskites using machine learning and first-principles calculations.

In this study, the machine-learning method, combined with density functional perturbation theory (DFPT) calculations, was implemented to predict and validate the dielectric constants of ABO3-type perovskites. For the construction of the training database, the dielectric constants of 7113 inorganic materials were extracted from the Materials Project. The chemical, structural, and physical descriptors were generated and trained using the gradient-boosting-based regressor after feature engineering. The prediction accuracies were 0.83 and 0.67 (R2) and 0.12 and 0.26 (root mean square error) for the electronic and ionic contributions to the dielectric constant, respectively. The constructed surrogate model was then employed to predict the dielectric constants of the ABO3-type perovskites (216 structures), whose thermodynamic stabilities were satisfactory. The predicted values were validated using DFPT calculations. The constructed database was further used to develop a surrogate model for the prediction of dielectric constants. The final R2 prediction accuracies reached 0.79 (electronic) and 0.67 (ionic).

A Comparative Study of the Performance for Predicting Biodegradability Classification: The Quantitative Structure-Activity Relationship Model vs the Graph Convolutional Network.

The prediction and evaluation of the biodegradability of molecules with computational methods are becoming increasingly important. Among the various methods, quantitative structure-activity relationship (QSAR) models have been demonstrated to predict the ready biodegradation of chemicals but have limited functionality owing to their complex implementation. In this study, we employ the graph convolutional network (GCN) method to overcome these issues. A biodegradability dataset from previous studies was trained to generate prediction models by (i) the QSAR models using the Mordred molecular descriptor calculator and MACCS molecular fingerprint and (ii) the GCN model using molecular graphs. The performance comparison of the methods confirms that the GCN model is more straightforward to implement and more stable; the specificity and sensitivity values are almost identical without specific descriptors or fingerprints. In addition, the performance of the models was further verified by randomly dividing the dataset into 100 different cases of training and test sets and by varying the test set ratio from 20 to 80%. The results of the current study clearly suggest the promise of the GCN model, which can be implemented straightforwardly and can replace conventional QSAR prediction models for various types and properties of molecules.

A Synergistic Effect of Na+ and Al3+ Dual Doping on Electrochemical Performance and Structural Stability of LiNi0.88Co0.08Mn0.04O2 Cathodes for Li-Ion Batteries.

The synergistic effect of Na+/Al3+ dual doping is investigated to improve the structural stability and electrochemical performance of LiNi0.88Co0.08Mn0.04O2 cathodes for Li-ion batteries. Rietveld refinement and density functional theory calculations confirm that Na+/Al3+ dual doping changes the lattice parameters of LiNi0.88Co0.08Mn0.04O2. The changes in the lattice parameters and degree of cation mixing can be alleviated by maintaining the thickness of the LiO6 slab because the energy of Al-O bonds is higher than that of transition metal (TM)-O bonds. Moreover, Na is an abundant and inexpensive metal, and unlike Al3+, Na+ can be doped into the Li slab. The ionic radius of Na+ (1.02 Å) is larger than that of Li+ (0.76 Å); therefore, when Na+ is inserted into Li sites, the Li slab expands, indicating that Na+ serves as a pillar ion for the Li diffusion pathway. Upon dual doping of the Li and TM sites of Ni-rich Ni0.88Co0.08Mn0.04O2 (NCM) with Na+ and Al3+, respectively, the lattice structure of the obtained NNCMA is more ideal than those of bare NCM and Li+- and Na+-doped NCM (NNCM and NCMA, respectively). This suggests that NNCMA with an ideal lattice structure presents several advantages, namely, excellent structural stability, a low degree of cation mixing, and favorable Li-ion diffusion. Consequently, the rate capability of NNCMA (83.67%, 3 C/0.2 C), which presents favorable Li-ion diffusion because of the expanded Li sites, is higher than those of bare NCM (78.68%), NNCM (81.15%), and NCMA (83.18%). The Rietveld refinement, differential capacity analysis, and galvanostatic intermittent titration technique results indicate that NNCMA exhibits low polarization, favorable Li-ion diffusion, and a low degree of cation mixing; moreover, its phase transition is hindered. Consequently, NNCMA demonstrates a higher capacity retention (84%) than bare NCM (79%), NNCM (82%), and NCMA (82%) after 50 cycles at 1 C. This study provides insight into the fabrication of Ni-rich NCMs with excellent electrochemical performance.

Adaptive Learning Framework in Prediction and Validation of Gibbs Free Energy for Inorganic Crystalline Solids.

Gibbs free energy is a fundamental physical property for understanding the stability and synthesizability of materials under various thermodynamic conditions, but its accessibility and availability are still limited. In this study, we used 9880 phonon databases to construct a machine learning model to predict approximately 40 000 Inorganic Crystalline Solid Database (ICSD) materials, whose free energy information has not been fully explored. To improve the prediction accuracy, a sampling strategy was implemented by including structures with low accuracy metrics, leading to R2 and mean absolute error values of 0.99 and 18.7 kJ/mol, respectively, in the testing set. Uncertainty analysis was followed for unexplored ICSD materials by obtaining the standard deviation in predictions from 10 surrogate models with different samplings in the training set. Based on this, an optimization process was conducted: density functional calculations were performed for 50 structures with high uncertainty and the training database was updated; this loop was repeated 15 times. This demonstrates the reduction and saturation in the uncertainty, confirming that the constructed model can provide a comprehensive map of the Gibbs free energy for inorganic solid materials. This can accelerate the material screening process by providing information on thermodynamic stability.

Searching for Mechanically Superior Solid-State Electrolytes in Li-Ion Batteries via Data-Driven Approaches.

Li-ion solid-state electrolytes (SSEs) have great potential, but their commercialization is limited due to interfacial contact stability issues and the formation and growth of dendrites. In this study, a machine learning regression algorithm was implemented to screen for mechanically superior SSEs among 17,619 candidates. Elasticity information (14,238 structures) was imported from an available database, and their machine learning descriptors were constructed using physiochemical and structural properties. A surrogate model for predicting the shear and bulk moduli exhibited R2 values of 0.819 and 0.863, respectively. The constructed model was applied to predict the elastic properties of potential SSEs, and first-principles calculations were conducted for validation. Furthermore, the application of an active learning process, which reduced the prediction uncertainty, was clearly demonstrated to improve the R2 score from approximately 0.6-0.8 by adding only 32-63% of new data sets depending on the type of modulus. We believe that the current model and additional data sets can accelerate the process of finding optimal SSEs to satisfy the mechanical conditions being sought.

Accelerated Discovery of Zeolite Structures with Superior Mechanical Properties via Active Learning.

A Bayesian active learning platform is developed for the accelerated discovery of mechanically superior zeolite structures from more than half a million hypothetical candidates. An initial database containing the mechanical properties of synthesizable zeolites is trained to develop the machine learning regression model. Then, a Bayesian optimization scheme is implemented to identify zeolites with potentially excellent mechanical properties. The newly accumulated database consists of 871 labeled structures, and the uncertainty of the predictive model is reduced by 40% and 58% for the bulk and shear moduli, respectively. The model convergence shows that no further improvement occurs after the 10th iteration of optimizations. The proposed platform is able to discover 23 new zeolite structures that have unprecedented shear moduli; in one case, the shear modulus (127.81 GPa) is 250% higher than the previous data set. The proposed platform accelerates the material discovery process while maximizing computational efficiency and enhancing the predictive accuracy.

Revealing the role of dopants in mitigating degradation phenomena in sodium-ion layered cathodes.

Prevention of the degradation of sodium-based layered cathode materials is the key to developing high-performance and high-stability sodium-ion batteries. In this study, the working mechanism of Mg and Ti dopants in mitigating degradation was investigated through the use of first-principles calculations. More specifically, the effects of each dopant in suppressing the phase transition, lattice expansion and shrinkage, and possible oxygen generation during repeated charging and discharging processes were validated. The results showed that the pristine structure exhibits irreversible O3-P3 phase transition after 75% desodiation, while doping with Mg or Ti effectively delays this transition. In addition, the change in lattice parameters as well as in the volume during desodiation was investigated. It was found that both dopants reduce the magnitude of structural change, which potentially improves the structural stability. Furthermore, introducing the dopants increases the thickness of the Na diffusion channel, possibly leading to an enhanced rate capability. Finally, the oxygen atomic charge variation during charging indicated that doping can enhance the oxygen stability by reducing the initial charge of oxygen as well as its increase during desodiation.