Functional ceramic materials database: an online resource for materials research

We present work on the creation of a ceramic materials database which contains data gleaned from literature data sets as well as new data obtained from combinatorial experiments on the London University Search Instrument. At the time of this writing, the database contains data related to two main groups of materials, mainly in the perovskite family. Permittivity measurements of electroceramic materials are the first area of interest, while ion diffusion measurements of oxygen ion conductors are the second. The nature of the database design does not restrict the type of measurements which can be stored; as the available data increase, the database may become a generic, publicly available ceramic materials resource.     

Optimization strategy based on genetic algorithms and neural networks applied to a polymerization process

The article presents a simple and general methodology, especially destined to the optimization of complex, strongly nonlinear systems, for which no extensive knowledge or precise models are available. The optimization problem is solved by means of a simple genetic algorithm, and the results are interpreted both from the mathematical point of view (the minimization of the objective function) and technological (the estimation of the achievement of individual objectives in multiobjective optimization). The use of a scalar objective function is supported by the fact that the genetic algorithm also computes the weights attached to the individual objectives along with the optimal values of the decision variables. The optimization strategy is accomplished in three stages: (1) the design and training of the neural model by a new method based on a genetic algorithm where information about the network is coded into the chromosomes; (2) the actual optimization based on genetic algorithms, which implies testing different values for parameters and different variants of the algorithm, computing the weights of the individual objectives and determining the optimal values for the decision variables; (3) the user's decision, who chooses a solution based on technological criteria. © 2007 Wiley Periodicals, Inc. Int J Quantum Chem, 2008     

Performance validation of neural network based (13)c NMR prediction using a publicly available data source

The validation of the performance of a neural network based 13C NMR prediction algorithm using a test set available from an open source publicly available database, NMRShiftDB, is described. The validation was performed using a version of the database containing ca. 214,000 chemical shifts as well as for two subsets of the database to compare performance when overlap with the training set is taken into account. The first subset contained ca. 93,000 chemical shifts that were absent from the ACD\CNMR DB, the "excluded shift set" used for training of the neural network and the ACD\CNMR prediction algorithm, while the second contained ca. 121,000 shifts that were present in the ACD\CNMR DB training set, the "included shift set". This work has shown that the mean error between experimental and predicted shifts for the entire database is 1.59 ppm, while the mean deviation for the subset with included shifts is 1.47 and 1.74 ppm for excluded shifts. Since similar work has been reported online for another algorithm we compared the results with the errors determined using Robien's CNMR Neural Network Predictor using the entire NMRShiftDB for program validation.     

Growing neural networks for a multivariate calibration and variable selection of time-resolved measurements

In this study, an algorithm for growing neural networks is proposed. Starting with an empty network the algorithm reduces the error of prediction by subsequently inserting connections and neurons. The type of network element and the location where to insert the element is determined by the maximum reduction of the error of prediction. The algorithm builds non-uniform neural networks without any constraints of size and complexity. The algorithm is additionally implemented into two frameworks, which use a data set limited in size very efficiently, resulting in a more reproducible variable selection and network topology.

The algorithm is applied to a data set of binary mixtures of the refrigerants R22 and R134a, which were measured by a surface plasmon resonance (SPR) device in a time-resolved mode. Compared with common static neural networks all implementations of the growing neural networks show better generalization abilities resulting in low relative errors of prediction of 0.75% for R22 and 1.18% for R134a using unknown data.     

A graph representation of molecular ensembles for polymer property prediction

Synthetic polymers are versatile and widely used materials. Similar to small organic molecules, a large chemical space of such materials is hypothetically accessible. Computational property prediction and virtual screening can accelerate polymer design by prioritizing candidates expected to have favorable properties. However, in contrast to organic molecules, polymers are often not well-defined single structures but an ensemble of similar molecules, which poses unique challenges to traditional chemical representations and machine learning approaches. Here, we introduce a graph representation of molecular ensembles and an associated graph neural network architecture that is tailored to polymer property prediction. We demonstrate that this approach captures critical features of polymeric materials, like chain architecture, monomer stoichiometry, and degree of polymerization, and achieves superior accuracy to off-the-shelf cheminformatics methodologies. While doing so, we built a dataset of simulated electron affinity and ionization potential values for >40k polymers with varying monomer composition, stoichiometry, and chain architecture, which may be used in the development of other tailored machine learning approaches. The dataset and machine learning models presented in this work pave the path toward new classes of algorithms for polymer informatics and, more broadly, introduce a framework for the modeling of molecular ensembles.

A graph representation that captures critical features of polymeric materials and an associated graph neural network achieve superior accuracy to off-the-shelf cheminformatics methodologies.     

Novel methods of automated structure elucidation based on 13C NMR spectroscopy

Three new approaches for automated structure elucidations of organic molecules using NMR spectroscopic data were introduced recently. These approaches apply a neural network 13C NMR chemical shift prediction method to rank the results of structure generators by their agreement of the predicted and experimental chemical shifts. These three existing implementations are compared using realistic example molecules. The applicability and reliability of such approaches is addressed.     

Data-driven machine learning models for quick prediction of thermal stability properties of OLED materials

Organic light-emitting diode (OLED) materials have exhibited a wide range of applications. However, the further development and commercialization of OLEDs requires higher quality OLED materials, including materials with a high thermal stability. Thermal stability is associated with the glass transition temperature (T_g) and decomposition temperature (T_d), but experimental determinations of these two important properties generally involve a time-consuming and laborious process. Thus, the development of a quick and accurate prediction tool is highly desirable. Motivated by the challenge, we explored machine learning (ML) by constructing a new dataset with more than 1,000 samples collected from a wide range of literature, through which ensemble learning models were explored. Models trained with the LightGBM algorithm exhibited the best prediction performance, where the values of mean absolute error, root mean squared error, and R² were 17.15 K, 24.63 K, and 0.77 for T_g prediction and 24.91 K, 33.88 K, and 0.78 for T_d prediction. The prediction performance and the generalization of the ML models were further tested by two applications, which also exhibited satisfactory results. Experimental validation further demonstrated the reliability and the practical potential of the ML-based models. In order to extend the practical application of the ML-based models, an online prediction platform was constructed. This platform includes the optimal prediction models and all the thermal stability data under study, and it is freely available at http://www.oledtppxmpugroup.com. We expect that this platform will become a useful tool for experimental investigation of T_g and T_d, accelerating the design of OLED materials with desired properties.     

Prediction dielectric constant of different ternary liquid mixtures at various temperatures and compositions using artificial neural networks

Artificial neural networks (ANNs) were successfully developed for the modeling and prediction dielectric constant of different ternary liquid mixtures at various temperatures (?10°C?≤?t?≤?80°C) and over the complete composition range (0?≤?x ₁,?x ₂,?x ₃?≤?1). A three-layered feed forward ANN with architecture 7-16-1 was generated using seven parameters as inputs and its output is dielectric constant of media. It was found that properly selected and trained neural network could fairly represent the dependence of dielectric constant of different ternary liquid mixtures on temperature and composition. For the evaluation of the predictive power of the generated ANN, an optimized network was applied for predicting the dielectric constant in the prediction set, which were not used in the modeling procedure. Squared correlation coefficient (R ²) and root mean square error for prediction set are 0.9997 and 0.2060, respectively. The mean percent deviation (MPD) for the property in the prediction set is 0.8892%. The results show nonlinear dependence of dielectric constant of ternary mixed solvent systems on temperature and composition is significant.     

Artificial neural system modeling of Monte Carlo simulations of polymers

In this work, a neural network was used to learn features in potential energy surfaces and relate those features to conformational properties of a series of polymers. Specifically, we modeled Monte Carlo simulations of 20 polymers in which we calculated the characteristic ratio and the temperature coefficient of the characteristic ratio for each polymer. We first created 20 rotational potential energy surfaces using MNDO procedures and then used these energy surfaces to produce 10000 chains, each chain 100 bonds long. From these results we calculated the mean-square end-to-end distance, the characteristic ratio and its corresponding temperature coefficient. A neural network was then used to model the results of these Monte Carlo calculations. We found that artificial neural network simulations were highly accurate in predicting the outcome of the Monte Carlo calculations for polymers for which it was not trained. The overall average error for prediction of the characteristic ratio was 4,82%, and the overall average error for prediction of the temperature coefficient was 0,89%.     

Influence of the counterion and co-ion diffusion coefficient values on some dielectric and electrokinetic properties of colloidal suspensions

The dependences of the conductivity increment, the electrophoretic mobility, and the permittivity increment on the counterion diffusion coefficient value were numerically determined. The use of the network simulation method made it possible to solve the governing equations for the whole range of counterion and co-ion diffusion coefficients and for very low frequencies, despite the far-reaching field-induced charge density outside the double layer. Calculations performed for different zeta potential and electrolyte concentration values show that increasing the counterion mobility, while keeping constant the electrolyte solution conductivity and the kappa a values, strongly increases the conductivity increment, barely affects the electrophoretic mobility, and strongly decreases the permittivity increment. The numerical results are discussed and compared to analytical predictions derived from the Shilov-Dukhin model, which generally leads to a good agreement, at least for high kappa a and moderate zeta.     

人工神经网络在溶解度预测方面的应用

溶解度作为一项重要的物化指标,一直是化学学科的研究重点。然而,通过实验测量获得数据耗时费力,因此,科研人员建立了多种理论方法来进行估算,其中,人工神经网络因其能够关联复杂的多变量情况而受到广泛关注。本文综述了人工神经网络在物质溶解度预测方面的应用,介绍了应用最广泛的3种神经网络(BP神经网络、小波神经网络、径向基神经网络)的模型结构、预测方法和预测优势,探讨了神经网络的不足以及改进方法。文章最后对神经网络在物质溶解度预测方面的发展前景进行了展望。与其他方法相比,人工神经网络技术在物质溶解度预测方面具有预测结果精确度高、操作简单等特点,具有广阔的应用前景,但输入变量选择、隐含层节点数确定、避免局部最优等问题还需逐步建立系统的理论指导。     

Modified feed-forward neural network structures and combined-function-derivative approximations incorporating exchange symmetry for potential energy surface fitting

The classical interchange (permutation) of atoms of similar identity does not have an effect on the overall potential energy. In this study, we present feed-forward neural network structures that provide permutation symmetry to the potential energy surfaces of molecules. The new feed-forward neural network structures are employed to fit the potential energy surfaces for two illustrative molecules, which are H(2)O and ClOOCl. Modifications are made to describe the symmetric interchange (permutation) of atoms of similar identity (or mathematically, the permutation of symmetric input parameters). The combined-function-derivative approximation algorithm (J. Chem. Phys. 2009, 130, 134101) is also implemented to fit the neural-network potential energy surfaces accurately. The combination of our symmetric neural networks and the function-derivative fitting effectively produces PES fits using fewer numbers of training data points. For H(2)O, only 282 configurations are employed as the training set; the testing root-mean-squared and mean-absolute energy errors are respectively reported as 0.0103 eV (0.236 kcal/mol) and 0.0078 eV (0.179 kcal/mol). In the ClOOCl case, 1693 configurations are required to construct the training set; the root-mean-squared and mean-absolute energy errors for the ClOOCl testing set are 0.0409 eV (0.943 kcal/mol) and 0.0269 eV (0.620 kcal/mol), respectively. Overall, we find good agreements between ab initio and NN prediction in term of energy and gradient errors, and conclude that the new feed-forward neural-network models advantageously describe the molecules with excellent accuracy.     

ProteinUnet—An efficient alternative to SPIDER3‐single for sequence‐based prediction of protein secondary structures

Predicting protein function and structure from sequence remains an unsolved problem in bioinformatics. The best performing methods rely heavily on evolutionary information from multiple sequence alignments, which means their accuracy deteriorates for sequences with a few homologs, and given the increasing sequence database sizes requires long computation times. Here, a single‐sequence‐based prediction method is presented, called ProteinUnet, leveraging an U‐Net convolutional network architecture. It is compared to SPIDER3‐Single model, based on long short‐term memory‐bidirectional recurrent neural networks architecture. Both methods achieve similar results for prediction of secondary structures (both three‐ and eight‐state), half‐sphere exposure, and contact number, but ProteinUnet has two times fewer parameters, 17 times shorter inference time, and can be trained 11 times faster. Moreover, ProteinUnet tends to be better for short sequences and residues with a low number of local contacts. Additionally, the method of loss weighting is presented as an effective way of increasing accuracy for rare secondary structures.     

Predicting coal ash fusion temperature based on its chemical composition using ACO-BP neural network

Coal ash fusion temperature is important to boiler designers and operators of power plants. Fusion temperature is determined by the chemical composition of coal ash, however, their relationships are not precisely known. A novel neural network, ACO-BP neural network, is used to model coal ash fusion temperature based on its chemical composition. Ant colony optimization (ACO) is an ecological system algorithm, which draws its inspiration from the foraging behavior of real ants. A three-layer network is designed with 10 hidden nodes. The oxide contents consist of the inputs of the network and the fusion temperature is the output. Data on 80 typical Chinese coal ash samples were used for training and testing. Results show that ACO-BP neural network can obtain better performance compared with empirical formulas and BP neural network. The well-trained neural network can be used as a useful tool to predict coal ash fusion temperature according to the oxide contents of the coal ash.     

二进小波神经网络用于示波计时电位测定

提出了二进小波神经网络的结构及算法，并用于单组分和多组分示波计时电位信号的浓度计算。在二进小波神经网络中选用了Morlet母小波和修理的误差反传前向神经网络。探讨了二进小波神经网络中的中小波基个数，初始学习速率因子和动量因子等参数对网络预测结果的影响。结果表明：二进小波神经网络对双组分和单组分示波计时电位信号中去极剂浓度的预测均有很好效果。