Search

A multitask deep neural network model was trained on more than 218k different glass compositions. This model, called GlassNet, can predict 85 different properties (such as optical, electrical, dielectric, mechanical, and thermal properties, as well as density, viscosity/relaxation, crystallization, surface tension, and liquidus temperature) of glasses and glass-forming liquids of different chemistries (such as oxides, chalcogenides, halides, and others). The model and the data used to train it are available in the GlassPy Python module as free and open source software for the community to use and build upon. As a proof of concept, GlassNet was used with the MYEGA viscosity equation to predict the temperature dependence of viscosity and outperformed another general purpose viscosity model available in the literature (ViscNet) on unseen data. An explainable AI algorithm (SHAP) was used to extract knowledge correlating the input (physicochemical information) and output (glass properties) of the model, providing valuable insights for glass manufacturing and design. It is hoped that GlassNet, with its free and open source nature, can be used to enable faster and better computer-aided design of new technological glasses. [ABSTRACT FROM AUTHOR]

In the development and optimization of chemical processes involving the selection of organic fluids, knowledge of the physical properties of compounds is vital. In many cases, it is complex to find experimental measurements for all substances, so it becomes necessary to have a tool to predict properties based on the characteristics of the molecule. One of the most extensively used methods in the literature is the estimation by contribution of functional groups, where properties are calculated using the constituent elements of the molecule. There are several models published in the literature, but they fail to represent a wide variety of compounds with high accuracy and simultaneously maintain a low computational complexity. The aim of this work is to develop a prediction model for eight thermodynamic properties (melting temperature, boiling temperature, critical pressure, critical temperature, critical volume, enthalpy of vaporization, enthalpy of fusion, and enthalpy of gas formation) based on the group contribution methodology by implementing a multilayer perceptron. Here, 2736 substances were used to train the neural network, whose prediction capacity was compared with other reference models available in the literature. The proposed model presents errors ranging from 1% to 5% for the different properties (except for the melting point), which improves the reference models with errors in the range of 3%–30%. Nevertheless, a difficulty in the prediction of the melting point is detected, which could represent an inherent hindrance to this methodology. [ABSTRACT FROM AUTHOR]

Membrane polymers are a promising technology for use in many challenging gas separation applications. The techniques of computer‐aided molecular design can be used to search through the massive molecular space of heteropolymers and develop a set of likely candidate repeat units matching specific physical property targets. However, reasonably accurate property prediction algorithms are needed, but these algorithms must be very fast in order to be combined with an optimization framework. Artificial neural networks (ANNs), a branch of machine learning, are applied in this work to predict the physical properties of polymers. All of the physical properties investigated were found to be predicted by ANNs with R2 scores exceeding 0.82. [ABSTRACT FROM AUTHOR]

Recently, the field of polymer nanocomposites has been an area of high scientific and industrial attention due to noteworthy improvements attained in these materials, arising from the synergetic combination of properties of a polymeric matrix and an organic or inorganic nanomaterial. The enhanced performance of those materials typically involves superior mechanical strength, toughness and stiffness, electrical and thermal conductivity, better flame retardancy and a higher barrier to moisture and gases. Nanocomposites can also display unique design possibilities, which provide exceptional advantages in developing multifunctional materials with desired properties for specific applications. On the other hand, machine learning (ML) has been recognized as a powerful predictive tool for data-driven multi-physical modelling, leading to unprecedented insights and an exploration of the system's properties beyond the capability of traditional computational and experimental analyses. This article aims to provide a brief overview of the most important findings related to the application of ML for the rational design of polymeric nanocomposites. Prediction, optimization, feature identification and uncertainty quantification are presented along with different ML algorithms used in the field of polymeric nanocomposites for property prediction, and selected examples are discussed. Finally, conclusions and future perspectives are highlighted. [ABSTRACT FROM AUTHOR]

• Combined structure descriptors and ANN for fast and accurate TiO 2 energy prediction. • ANN shows limited prediction for band gap values with intense data fluctuation. • Key factors affecting stability and photoelectronic performance are identified. • Increasing complexity of ANN model does not necessarily boost predictive power. • Structural factors contributing to variations in TiO 2 band gap are more complicated. Fast and accurate prediction of certain properties from a given structure, as well as pinpointing key factors influencing particular properties, is crucial for the targeted design of new compounds with desired properties or crystal phase compositions. In the current work, a total of 2,419,200 Gaussian-type structure descriptors (GTSDs) were utilized as a bridge to establish artificial neural network (ANN) potential models between 1400 titanium dioxide (TiO 2) structures and their corresponding system energies and band gaps. After training, the ANN showed high accuracy in predicting energy, with R train of 0.985, R test of 0.967, and RMSE of 0.0489. However, for the band gap values with intense data fluctuation, the R train , R test , and RMSE changed to 0.864, 0.832, and 0.1304, respectively, indicating that the ANN's predictive ability for drastically changing data is limited. Compared with ab initio calculations, using ANN model can increase the calculation speed by 4–6 orders of magnitude. Furthermore, we identified the most significant features affecting the system energy and band gap values as the shape and distortion of [TiO x ] polyhedra through the random forest model. Our work aims to provide a new toolkit and research ideas for accelerating the development of new materials. [Display omitted] [ABSTRACT FROM AUTHOR]

Artificial intelligence (AI) has been referred to as the "fourth paradigm of science," and as part of a coherent toolbox of data‐driven approaches, machine learning (ML) dramatically accelerates the computational discoveries. As the machinery for ML algorithms matures, significant advances have been made not only by the mainstream AI researchers, but also those work in computational materials science. The number of ML and artificial neural network (ANN) applications in the computational materials science is growing at an astounding rate. This perspective briefly reviews the state‐of‐the‐art progress in some supervised and unsupervised methods with their respective applications. The characteristics of primary ML and ANN algorithms are first described. Then, the most critical applications of AI in computational materials science such as empirical interatomic potential development, ML‐based potential, property predictions, and molecular discoveries using generative adversarial networks (GAN) are comprehensively reviewed. The central ideas underlying these ML applications are discussed, and future directions for integrating ML with computational materials science are given. Finally, a discussion on the applicability and limitations of current ML techniques and the remaining challenges are summarized. This article is categorized under:Computer and Information Science > Chemoinformatics.Structure and Mechanism > Computational Materials Science.Computer and Information Science > Computer Algorithms and Programming.Software > Molecular Modeling. [ABSTRACT FROM AUTHOR]

Graph neural networks (GNNs) have been widely used for predicting properties and discovering structure–property relationships in chemistry and drug discovery. However, current GNNs treat all atoms as independent of each other, neglecting the crucial role of functional groups in molecules. In this work, graph neural networks based on molecular segmentation are proposed. An unsupervised segmentation method is proposed to partition molecular graph data into multiple functional group-based clusters. Segmentation message passing neural network is proposed to learn function groups, which generate embeddings in and between molecule clusters. To explain structure–property relationships, we propose a new explainer to identify substructures that are more compatible with the chemical principles analysis. Our approach attempts to train and explain GNNs more rationally by segmenting molecules. Experimental results show that our approach achieves more efficient and rational performance prediction and explanation on chemical and drug datasets. [Display omitted] • Graph neural networks are designed to learn important functional groups. • Higher prediction of molecular properties with lower time cost. • Reasonable molecule structure–property relationships results. • Unsupervised molecular segmentation without prior knowledge. [ABSTRACT FROM AUTHOR]

Abstract The glass transition temperature (T g) is a kinetic property of major importance for both fundamental and applied glass science. In this study, we designed and trained an artificial neural network to induce a model that can predict the T g of multicomponent oxide glasses. To do this, we used a dataset containing more than 55,000 inorganic glass compositions and their respective experimental values of T g. These compositions contain from 3 to 21 of the 45 chemical elements studied here. We implemented an optimization procedure to find artificial neural network hyperparameter values that were able to induce a model with high predictive performance. The resulting neural network model can correctly predict, with 95% accuracy, the published T g value within less than ±9% error, whereas 90% of the data are predicted with a relative deviation lower than ±6%. This level of uncertainty is equivalent to the level present in the original dataset and allows a very satisfactory description of the T g for multicomponent oxide glasses containing combinations of the 45 studied chemical elements. The prediction uncertainty does not depend on the number of elements in the glass composition. However, it is larger for glasses having very high T g (above 1250 K). The most important aspect is the algorithm's ability to predict the T g of glasses that are not included in the experimental dataset used for training, thus showing a high generalization ability. Besides, the procedure used here is general and can be easily extended to predict several other properties as a function of the glass composition. This handy feature will most probably help to develop new multicomponent glass compositions having remarkable properties. Graphical abstract Image 1 [ABSTRACT FROM AUTHOR]

An artificial neuron network based on genetic algorithm is presented to predict the normal boiling point (T b ) of refrigerants from 16 molecular groups and a topological index. The 16 molecular groups used in this paper can cover most refrigerants or working fluids in refrigeration, heat pump and organic Rankine cycle; the chosen topological index is able to distinguish all the refrigerant isomers. A total of 334 data points from previous experiments are used to create this network. The calculated results, which are based on a developed numerical method, show a good agreement with experimental data; the average absolute deviations for training, validation and test sets are 1.83%, 1.77%, 2.13%, respectively. A performance comparison between the developed numerical model and the other two existing models, namely QSPR approach and UNIFAC group contribution method, shows that the proposed model can predict T b of refrigerants in a better accord with experimental data. [ABSTRACT FROM AUTHOR]

[Display omitted] • A material recommendation system based on DeepFM was developed to predict the missing values in the MOFs material-property matrix. • 28 different properties of MOFs were extracted and used in DeepFM model. • The developed material recommendation system has good cold-start resistance with limited available data. • The model successfully predicted top 7 MOFs with highest CO 2 uptake and CO 2 /N 2 selectivity out of 8206 MOFs. • The model also offers the relative importance of various descriptors. Metal-organic frameworks (MOFs) has been widely considered as promising candidates for CO 2 adsorption due to their high porosity and high structural adjustability. By combining the various properties of MOFs, an MOFs material-property matrix can be obtained to find the best MOFs for different applications, but this matrix is often incomplete in practice. In this work, a DeepFM model was developed to predict the multi properties of CO 2 absorbents with limited start-up training data, but with high prediction accuracy. The DeepFM model contains deep neural network, with better non-linear fitting ability, thus significantly improved the prediction accuracy. Meanwhile, by adding the descriptors of MOFs as the input data of new features, the DM model also alleviates the cold start problem. By predicting 28 adsorption properties of 8206 screened hypothetical MOFs, 7 high-performance MOFs for CO 2 adsorption were selected. In addition, the model also helps to find out the relative importance degree of various descriptors on the CO 2 capture capabilities of MOFs, which is of great help in future MOFs synthesis. [ABSTRACT FROM AUTHOR]

The feasibility of using cascade neural networks to correlate the bubble points of ternary mixtures containing ionic liquids (ILs) and classical solvents was investigated. The systems investigated consisted of the ILs from the alkylimidazolium family with either the chloride, tetrafluoroborate, methylsulfate, or ethyl sulfate anions. The classical solvents investigated included 1-propanol, 2-propanol, ethanol, ethyl ethanoate, and water. A total of 272 bubble points were used in the training (205 data points) and testing (67 data points) stages. The optimized network comprised of nine neurons in the hidden layer and used the tangent-sigmoid and purelin functions as the activation functions in the hidden and output layers, respectively. This proposed network was able to correlate the ternary bubble points of both the training (AARD % = 0.13% and R2 = 0.9921) and the testing (AARD % = 0.15% and R2 = 0.9873) subsets with good accuracy, revealing the good correlative capability of the network. The statistical error analysis of all the data indicated an overall absolute average relative deviation (AARD %), mean square error (MSE) and correlation coefficient (R2) of 0.13%, 0.44% and 0.9909%, respectively. Predicted bubble points versus the experimental bubble points for three different ternary systems. [ABSTRACT FROM AUTHOR]

The data on the physical properties of steels which depend on temperature are needed for the calculation and simulation of heating and cooling processes. A method for predicting thermal conductivity of steels at elevated temperatures (up to 700 °C) from the known steel chemical composition has been developed, and the results obtained by the simulation are given. A static multi-layer feed-forward artificial neural network with the back propagation training function and Levenberg-Marquardt optimization was used to predict the coefficient of thermal conductivity of steels. In order to prevent the over fitting the early stopping method was applied. The following groups of steel were included in the model: structural steels, hotwork tool steels, high-speed steels, stainless steels, heat resistant steels austenitic steels for elevated temperatures, and cobalt alloyed steels and alloys for elevated temperatures. The mean absolute error in predicting thermal conductivity and the standard deviation were found to be very acceptable. [ABSTRACT FROM AUTHOR]

[Display omitted] • The structure–function relationships between molecular structures of DESs and physicochemical properties were analyzed. • A guidance about the proper structures of HBDs and HBAs for DESs with specific physicochemical properties was provided. • Effective and convenient neural network model systems were constructed for the prediction of DESs properties. Deep eutectic solvents (DESs) have been recognized as environmentally friendly alternatives to the conventional fossil solvents, which are in consistent with the concept of green chemistry. In order to offer theoretical guidance for designing DESs with specific physicochemical properties, the structure–function relationships between the molecular structures of DESs components and their physicochemical properties were systematically analyzed. The investigation of the functional groups, carbon chain length and type of halogen ions in DESs components provided a deeper understanding of the formation and properties of DESs. Density, viscosity, solvatochromic parameters (E NR value、π* value、α value and β value) were selected as the analyzable physicochemical properties. Except for the conclusion about the proper hydrogen bond donors (HBDs) and hydrogen bond acceptors (HBAs) for DESs with specific physicochemical properties, an effective neural network model system was also provided for the prediction of DESs properties. The molecular structural parameters of DESs used as input layers were calculated from the initial and constant molecular structural parameters of HBAs and HBDs. Not only considerable time was saved without complex calculation of the interactions energy between components, it was also helpful in constructing comprehensive database of the molecular information of the DESs components to provide convenience for future research. [ABSTRACT FROM AUTHOR]

The present paper describes various classification techniques like cluster analysis, principal component (PC)/factor analysis to classify different types of base stocks. The API classification of base oils (Group I–III) has been compared to a more detailed NMR derived chemical compositional and molecular structural parameters based classification in order to point out the similarities of the base oils in the same group and the differences between the oils placed in different groups. The detailed compositional parameters have been generated using 1H and 13C nuclear magnetic resonance (NMR) spectroscopic methods. Further, oxidation stability, measured in terms of rotating bomb oxidation test (RBOT) life, of non-conventional base stocks and their blends with conventional base stocks, has been quantitatively correlated with their 1H NMR and elemental (sulphur and nitrogen) data with the help of multiple linear regression (MLR) and artificial neural networks (ANN) techniques. The MLR based model developed using NMR and elemental data showed a high correlation between the ‘measured’ and ‘estimated’ RBOT values for both training (R=0.859) and validation (R=0.880) data sets. The ANN based model, developed using fewer number of input variables (only 1H NMR data) also showed high correlation between the ‘measured’ and ‘estimated’ RBOT values for training (R=0.881), validation (R=0.860) and test (R=0.955) data sets. [Copyright &y& Elsevier]

• Tensile properties of 18Cr-12Ni-Mo steels were modeled using neural networks. • Graphical user interface (GUI) was developed for easy use. • The model was able to correlate the relationships between input-output variables. • A virtual 18Cr-12Ni-Mo steels were created with mean values of the databases. • Effect of temperature and composition on properties was estimated accurately. An artificial neural network (ANN) model was designed to correlate the complex relations among composition, temperature, and mechanical properties of 18Cr-12Ni-Mo austenitic stainless steels. The developed model was used to estimate the composition-property and temperature-property correlations with 97% and 91% accuracy, for train and unseen test datasets. The ANN predictions are more accurate with experimental results as compared with the calculated properties of the existing model. The effective response of the alloying elements on the mechanical properties at ambient as well as elevated temperatures was quantitatively estimated with the help of the index of relative importance (I RI). The calculated results of the ANN model beneficial for both researchers as well as designers to guide actual experiments. Hence, this proposed technique will be helpful in developing the components of austenitic stainless steel with desired properties. [ABSTRACT FROM AUTHOR]

Artificial Intelligence (AI) plays a pivotal role in drug discovery. In particular artificial neural networks such as deep neural networks or recurrent networks drive this area. Numerous applications in property or activity predictions like physicochemical and ADMET properties have recently appeared and underpin the strength of this technology in quantitative structure-property relationships (QSPR) or quantitative structure-activity relationships (QSAR). Artificial intelligence in de novo design drives the generation of meaningful new biologically active molecules towards desired properties. Several examples establish the strength of artificial intelligence in this field. Combination with synthesis planning and ease of synthesis is feasible and more and more automated drug discovery by computers is expected in the near future. [ABSTRACT FROM AUTHOR]