Neural networks for Nyquist plots prediction during corrosion inhibition of a pipeline steel

This paper presents a predictive model for electrochemical impedance Nyquist plots using artificial neural network. The proposed model obtains predictions of imaginary impedance based on the real part of the impedance as a function of time. The model takes into account the variations of the real impedance and immersion time of steel in a corrosive environment, considering constant carboxyamido-imidazoline corrosion inhibitor concentrations (5 and 25 ppm). For the network, the Levenberg–Marquardt learning algorithm, the hyperbolic tangent sigmoid transfer function, and the linear transfer function were used. The best-fitting training data set was obtained with five neurons in the hidden layer for 5 ppm of inhibitor and two neurons in the hidden layer for 25 ppm of inhibitor, which made it possible to predict the efficiency with accuracy at least as good as that of the theoretical error, over the whole theoretical range. On the validation data set, simulations and theoretical data test were in good agreement with an R value of 0.984 for 5 ppm and 0.994 for 25 ppm of inhibitor. The developed model can be used for the prediction of the real and imaginary parts of the impedance as a function of time for short simulation times.     

Artificial neural network for the evaluation of CO2 corrosion in a pipeline steel

This paper presents a predictive model for the determination of different types of corrosion by using electrochemical impedance spectroscopy curves and artificial neural network. This proposed model obtains predictions for three different types of corrosion by using Nyquist impedance curves from four input variables: inhibitor concentration, time of exposure, and the real and imaginary experimental component of these curves. The model takes into account the variations of inhibitor concentration over steel to decrease the corrosion rate. For the network, the Levenberg–Marquardt learning algorithm, the hyperbolic tangent sigmoid transfer function and the linear transfer function were used. The best fitting training data set was obtained with five neurons in the hidden layer, which made possible to predict satisfactory efficiency (R > 0.99). On the validation of the data set, simulations and theoretical data tests were in good agreement (R > 0.9905). The developed model can be used for the determination of the type of curves related to the nature phenomena and rate of corrosion at the metal surface.     

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.     

Neural Network Classification of Mutagens Using Structural Fragment Data

Abstract

A neural network was applied to a large, structurally heterogeneous data set of mutagens and nonmutagens to investigate structure-property relationships. Substructural data comprising a total of 1280 fragments were used as inputs. The training of the back-propagation networks was directed by an algorithm which selected an optimal subset of fragments in order to maximize their discriminating power, and a good predictive network.

The system comprised three programs: the first used a keyfile of 100 fragments to generate training and test files, the second was the network itself and a procedure for ranking the effectiveness of these fragments and the third randomly replaced the lowest fragments. This cycle was then repeated. After running on a 386/33 PC several networks produced approximately 11% failures in the test set and 6% in the training set.

By simplifying the output of the hidden layer it was possible to describe the hidden layer states in terms of clusters of mutagens and non-mutagens. Some of these clusters were structurally homogeneous and contained known mutagenic and non-mutagenic structural classes. This analysis provided a useful means of demonstrating how the network was classifying the data.     

Simulation studies of the fidelity of biomolecular structure ensemble recreation

We examine the ability of Bayesian methods to recreate structural ensembles for partially folded molecules from averaged data. Specifically we test the ability of various algorithms to recreate different transition state ensembles for folding proteins using a multiple replica simulation algorithm using input from "gold standard" reference ensembles that were first generated with a Go-like Hamiltonian having nonpairwise additive terms. A set of low resolution data, which function as the "experimental" phi values, were first constructed from this reference ensemble. The resulting phi values were then treated as one would treat laboratory experimental data and were used as input in the replica reconstruction algorithm. The resulting ensembles of structures obtained by the replica algorithm were compared to the gold standard reference ensemble, from which those "data" were, in fact, obtained. It is found that for a unimodal transition state ensemble with a low barrier, the multiple replica algorithm does recreate the reference ensemble fairly successfully when no experimental error is assumed. The Kolmogorov-Smirnov test as well as principal component analysis show that the overlap of the recovered and reference ensembles is significantly enhanced when multiple replicas are used. Reduction of the multiple replica ensembles by clustering successfully yields subensembles with close similarity to the reference ensembles. On the other hand, for a high barrier transition state with two distinct transition state ensembles, the single replica algorithm only samples a few structures of one of the reference ensemble basins. This is due to the fact that the phi values are intrinsically ensemble averaged quantities. The replica algorithm with multiple copies does sample both reference ensemble basins. In contrast to the single replica case, the multiple replicas are constrained to reproduce the average phi values, but allow fluctuations in phi for each individual copy. These fluctuations facilitate a more faithful sampling of the reference ensemble basins. Finally, we test how robustly the reconstruction algorithm can function by introducing errors in phi comparable in magnitude to those suggested by some authors. In this circumstance we observe that the chances of ensemble recovery with the replica algorithm are poor using a single replica, but are improved when multiple copies are used. A multimodal transition state ensemble, however, turns out to be more sensitive to large errors in phi (if appropriately gauged) and attempts at successful recreation of the reference ensemble with simple replica algorithms can fall short.     

Estimation of aqueous solubility for a diverse set of organic compounds based on molecular topology

An accurate and generally applicable method for estimating aqueous solubilities for a diverse set of 1297 organic compounds based on multilinear regression and artificial neural network modeling was developed. Molecular connectivity, shape, and atom-type electrotopological state (E-state) indices were used as structural parameters. The data set was divided into a training set of 884 compounds and a randomly chosen test set of 413 compounds. The structural parameters in a 30-12-1 artificial neural network included 24 atom-type E-state indices and six other topological indices, and for the test set, a predictive r2 = 0.92 and s = 0.60 were achieved. With the same parameters the statistics in the multilinear regression were r2 = 0.88 and s = 0.71, respectively.     

Development of novel 3D-QSAR combination approach for screening and optimizing B-Raf inhibitors in silico

B-Raf is a member of the RAF family of serine/threonine kinases: it mediates cell division, differentiation, and apoptosis signals through the RAS-RAF-MAPK pathway. Thus, B-Raf is of keen interest in cancer therapy, such as melanoma. In this study, we propose the first combination approach to integrate the pharmacophore (PhModel), CoMFA, and CoMSIA models for B-Raf, and this approach could be used for screening and optimizing potential B-Raf inhibitors in silico. Ten PhModels were generated based on the HypoGen BEST algorithm with the flexible fit method and diverse inhibitor structures. Each PhModel was designated to the alignment rule and screening interface for CoMFA and CoMSIA models. Therefore, CoMFA and CoMSIA models could align and recognize diverse inhibitor structures. We used two quality validation methods to test the predication accuracy of these combination models. In the previously proposed combination approaches, they have a common factor in that the number of training set inhibitors is greater than that of testing set inhibitors. In our study, the 189 known diverse series B-Raf inhibitors, which are 7-fold the number of training set inhibitors, were used as a testing set in the partial least-squares validation. The best validation results were made by the CoMFA09 and CoMSIA09 models based on the Hypo09 alignment model. The predictive r(2)(pred) values of 0.56 and 0.56 were derived from the CoMFA09 and CoMSIA09 models, respectively. The CoMFA09 and CoMSIA09 models also had a satisfied predication accuracy of 77.78% and 80%, and the goodness of hit test score of 0.675 and 0.699, respectively. These results indicate that our combination approach could effectively identify diverse B-Raf inhibitors and predict the activity.     

The impact of available experimental data on the prediction of 1H NMR chemical shifts by neural networks

Two different ways were explored to incorporate new available experimental data into previously trained ensembles of feed-forward neural networks, for the structure-based prediction of (1)H NMR chemical shifts of organic compounds. One approach used the new data as the memory of an associative neural network (ASNN) system. For an independent prediction set of 952 cases, a mean average error of 0.19 ppm was achieved (0.13 ppm for 90% of the cases). This approach advantageously avoids retraining the networks, and the predictions compared favorably with those obtained by available commercial software packages. Excellent predictions could also be achieved by retraining the networks with the new data, but only if the training sets were selected so as to be balanced or if the retraining started with the weights of the previously trained networks.