Bayesian Regularized Artificial Neural Network Research Articles

Robust data-driven model to study dispersion of vapor cloud in offshore facility

Data driven models are increasingly used in engineering design and analysis. Bayesian Regularization Artificial Neural Network (BRANN) and Levenberg-Marquardt Artificial Neural Network (LMANN) are two widely used data-driven models. However, their application to study the dispersion in complex geometry is not explored.This study aims to investigate the suitability of BRANN and LMANN in estimating dimension of flammable cloud in congested offshore platform. A large number of numerical simulations are conducted using FLACS. Part of these simulations results are used to training the network. The trained network is subsequently used to predict the vapor cloud dimension and compared against remaining simulation results. The predictive abilities of these network along with Response Surface Method and Frozen Cloud Approach (FCA) are studied. The comparative results indicate BRANN model with 20 hidden neurons is the most robust and precise. The developed BRANN would serve an effective and tool for quick Explosion Risk Analysis ERA.

Optimal Configuration of Vortex Generator for Heat Transfer Enhancement in a Plate-Fin Channel

Heat transfer is a naturally occurring phenomenon and its augmentation is a vital research topic for many years. Although, vortex generators (VGs) are widely used to enhance the heat transfer of plate-fin type heat exchangers, few researches deal with its thermal optimization. This work is dedicated to the numerical investigation and optimization of VGs configuration in a plate-fin channel. Three-dimensional (3D) numerical simulations are performed to study the effect of angle of attack and attach angle (angle between VG and wall) and shape of VG on the fluid flow and heat transfer characteristics. The flow is assumed as steady-state, incompressible, and laminar within the range of studied Reynolds numbers (Re = 380–1140). Results are presented in the form average and local Nusselt number and friction factor. The effect of attach angle is highlighted and the results show that the attach angle of 90 deg may not be necessary for enhancing the heat transfer. The flow structure and heat transfer characteristics of certain cases are examined in detail. The parameters of VG are then optimized for maximum heat transfer and minimum pressure drop. The three independent design parameters are considered for the two objective functions. For this purpose, computation fluid dynamics (CFD) data, response surface methodology (RSM) and a multi-objective optimization algorithm (MOA) are combined. The data obtained from numerical simulations are used to train a Bayesian-regularized artificial neural network (BRANN). This in turn is used to drive a MOA to find the optimal parameters of VGs in the form of Pareto front. The optimal values of these parameters are finally presented.

Application of bayesian regularized artificial neural networks to predict pour point of crude oil treated by pour point depressant

ABSTRACTThe pour point of the crude oil treated with the pour point depressant (PPD) is easily affected by the shear history effect. Models for pour point of PPD-treatment crude oil affected by the shear history effect based on Bayesian regularized artificial neural network (BRANN) were established. The results showed that BRANN models not only had a good ability of fitting to the training data, but also had a good ability of predicting the testing data. By evaluating network performance with several statistical indicators, the three models have excellent performance, high accuracy, and strong generalization ability. The influence of each parameter on the pour point were also investigated through a sensitivity analysis, which shows that the entropy generation due to viscous flow is the most important parameter in predicting the pour point.

Magnetic metal-organic frameworks for the extraction of trace amounts of heavy metal ions prior to their determination by ICP-AES

The authors describe the preparation of two kinds of metal-organic frameworks (MOFs), referred to as TMU-8 and TMU-9. The MOFs were applied to the preconcentration of the ions Co(II), Cu(II), Pb(II), Cd(II), Ni(II), Cr(III), and Mn(II) from aqueous solutions. The roles of the azine groups in TMU-8 (in comparison to TMU-9 which does not have an azine group) and the role of void spaces of these MOFs toward the adsorption of metal ions also are evaluated. The studies reveal that TMU-8 has a better adsorption capability than TMU-9. A magnetic TMU-8 was then fabricated by in-situ synthesis of a magnetic core-shell nanocomposite. The material was chosen as an efficient sorbent for the preconcentration of the above metal ions, followed by their determination by flow injection inductively coupled plasma atomic emission spectrometry. The assay was optimized using a combination of central composite design (CCD) and a Bayesian regularized artificial neural network (BRANN) technique. Under optimal conditions, the preconcentration factors are in the range between 66 and 232, and detection limits are as low as 0.3 to 1 μg ⋅L−1. The relative standard deviations are <6.4% (for n = 3; at 50 μg ⋅ L−1). Real samples were analyzed, and the results demonstrate that such core-shell magnetic microspheres are promising sorbents for rapid and efficient extraction of heavy metal ions from complex samples.

A laboratory study on mix design to properly resemble a jointed brittle rock

In this paper attempts have been done to create a mortar with relatively high uniaxial compressive strength (UCS), easy casting, high flexibility, instant hardening, low cost and easy availability. The main use of this material is to physically model the mechanical behavior of jointed rock-like blocks. The effect of four parameters such as joint roughness coefficient (JRC), bridge length (L), bridge angle (γ) and joint inclination (θ) on UCS of non-persistent jointed blocks were studied. For this purpose, 35 cylindrical specimens with a broad range of plaster content (P) and cement content (C) in different ages were tested. In order to increase the strength of blocky specimens, some retarder and lubricant were used. The results showed that using 3 wt. % (Weight percent) lubricant MGAR106 and 0.05 wt. % Retarder decreases water content by 12.5% and increases plaster and cement content of 8.3% and 4.17 % respectively. Consequently, UCS of blocky specimens increased by 284.33%. In order to formulize the effect of P/C content and the age of cylindrical specimens (A) on UCS, Multivariate Non-linear Regression (MNR) and Bayesian Regularized Artificial Neural Network (BRANN) models were deployed. The results showed that BRANN approach can provide more exact predictions of the specimen UCS than MNR model. Moreover, P/C content had more influence on UCS than the specimen age. Finally the UCS tests on blocky specimens indicated that increasing JRC, bridge length and bridge angle increases UCS and it takes its minimum ate joint inclination of 60°. Furthermore, the capability of produced material to model cracking behaviour of jointed blocks was approved.

Synthesis of Fe3O4@PPy–MWCNT nanocomposite and its application for extraction of ultra-trace amounts of PAHs from various samples

In the present study, multi-walled carbon nanotube oxide was immobilized on the pyrrole magnetic nanoparticles. Application of the synthesized material was investigated for the magnetic solid-phase extraction (MSPE) of polycyclic aromatic hydrocarbons (PAHs), from the environmental samples. Determinations of the analytes were performed with gas chromatography–mass spectrometry. The structure and morphology of Fe3O4@PPy–MWCNT were characterized by Fourier transform infrared spectroscopy, scanning electron microscopy, thermal gravimetric analysis, and vibrating sample magnetometer. Performance of MSPE is mainly affected by extraction time, sorbent amount, sample solution volume, and eluent type and volume. In this study, the best possible performance of MSPE has been achieved using a combination of central composite design and Bayesian regularized artificial neural network technique. Under the optimum extraction conditions, linear range between 0.5 and 250 µg L−1 (R 2 > 0.994), preconcentration factors from 232 to 403 and limits of detection ranging from 0.1 to 0.3 µg L−1 were obtained. Relative standard deviations for intra-day and inter-day precision were 3.3–5.1% and 3.7–5.6%, respectively. In addition, feasibility of the method was demonstrated by extraction and determination of PAHs from some real samples containing tap water, hookah water as well as soil samples, and relative recovery in the range of 85.4–106.8% was obtained. This MSPE method provides several advantages, such as high extraction efficiency, minimum sorbent for extraction of the analytes from high sample volumes, convenient extraction procedure, and short analysis times.

Accessing marker effects and heritability estimates from genome prediction by Bayesian regularized neural networks

Recently, there is an increasing interest on semi- and non-parametric methods for genome-enabled prediction, among which the Bayesian regularized artificial neural networks (BRANN) stand. We aimed to evaluate the predictive performance of BRANN and to exploit SNP effects and heritability estimates using two different approaches (relative importance-RI, and relative contribution-RC). Additionally, we aimed also to compare BRANN with the traditional RR-BLUP and BLASSO by using simulated datasets. The simplest BRANN (net1), RR-BLUP and BLASSO methods outperformed other more parameterized BRANN (net2, net3, … net6) in terms of predictive ability. For both simulated traits (Y1 and Y2) the net1 provided the best h2 estimates (0.33 for both, being the true h2=0.35), whereas RR-BLUP (0.18 and 0.22 for Y1 and Y2, respectively) and BLASSO (0.20 and 0.26 for Y1 and Y2, respectively) underestimated h2. The marker effects estimated from net1 (using RI and RC approaches) and RR-BLUP were similar, but the shrinkage strength was remarkable for BLASSO on both traits. For Y1, the correlation between the true fifty QTL effects and the effects estimated for the SNPs located in the same QTL positions were 0.61, 0.60, 0.60 and 0.55, for RI, RC, RR-BLUP and BLASSO; and for Y2, these correlations were 0.81, 0.81, 0.81 and 0.71, respectively. In summary, we believe that estimates of SNP effects are promising quantitative tools to bring discussions on chromosome regions contributing most effectively to the phenotype expression when using ANN for genomic predictions.

Predictive Abilities of Bayesian Regularization and Levenberg–Marquardt Algorithms in Artificial Neural Networks: A Comparative Empirical Study on Social Data

The objective of this study is to compare the predictive ability of Bayesian regularization with Levenberg–Marquardt Artificial Neural Networks. To examine the best architecture of neural networks, the model was tested with one-, two-, three-, four-, and five-neuron architectures, respectively. MATLAB (2011a) was used for analyzing the Bayesian regularization and Levenberg–Marquardt learning algorithms. It is concluded that the Bayesian regularization training algorithm shows better performance than the Levenberg–Marquardt algorithm. The advantage of a Bayesian regularization artificial neural network is its ability to reveal potentially complex relationships, meaning it can be used in quantitative studies to provide a robust model.

Heat transfer and optimization studies on layered porous stackings under an imposed pressure drop

This work reports the results of a study to augment the performance of layered porous heat sinks under forced or mixed convection. A two dimensional computational domain is used in the study which involves a rectangular channel consisting of a layered porous stacking. The temperature and flow fields are obtained for different configurations and inlet velocities with the Local Thermal Equilibrium (LTE) model using the commercially available software Fluent 14.0. The solver was validated with in-house experiments and the predictions from the solver were found to be in good agreement with the experimental results. The data obtained from the numerical experiments were used to train a BRANN (Bayesian-regularized artificial neural network). This, in turn, is used to drive a genetic algorithm (GA) to determine the optimal porosity distribution across the layers and the inlet velocity, maximizing the heat transfer with the simultaneous consideration of minimizing the pressure drop. The study reveals that factoring in the pressure drop during design results in a deviation from the optimal configuration obtained by minimizing only the hot-spot temperature. The optimal configuration has been determined and the benefit of an appropriate porosity distribution on the thermal performance has been investigated, giving qualitative insights into the performance of layered porous media.

A robust framework to predict mercury speciation in combustion flue gases

Mercury emissions from coal combustion have become a global concern as growing energy demands have increased the consumption of coal. The effective implementation of treatment technologies requires knowledge of mercury speciation in the flue gas, namely concentrations of elemental, oxidized and particulate mercury at the exit of the boiler. A model that can accurately predict mercury species in flue gas would be very useful in that context. Here, a Bayesian regularized artificial neural network (BRANN) that uses five coal properties and combustion temperature was developed to predict mercury speciation in flue gases before treatment technology implementation. The results of the model show that up to 97 percent of the variation in mercury species concentration is captured through the use of BRANNs. The BRANN model was used to conduct a parametric sensitivity which revealed that the coal chlorine content and coal calorific value were the most sensitive parameters, followed by the combustion temperature. The coal sulfur content was the least important parameter. The results demonstrate the applicability of BRANNs for predicting mercury concentration and speciation in combustion flue gas and provide a more efficient and effective technique when compared to other advanced non-mechanistic modeling strategies.

QSAR study of necroptosis inhibitory activities (EC50) of [1,2,3] thiadiazole and thiophene derivatives using Bayesian regularized artificial neural network and calculated descriptors

A Bayesian regularized artificial neural network (BR-ANN) model was developed for modeling and accurate prediction of necroptosis inhibitory activities of [1,2,3] thiadiazole and thiophene derivatives as potent tumor necrosis factor-α-induced necroptosis inhibitors. The chemical structure of each compound was converted to 1,481 molecular descriptors by Dragon software. Among these, only seven descriptors relating the activity data to the molecular structures were selected as the significant ones. The best BR-ANN model was a three layer feed forward network with the 7-4-1 architecture. Prediction ability of the proposed model was evaluated by prediction of pEC50 of some compounds in the external (test) data set. The mean square error, mean absolute error, correlation coefficient (R), and mean relative error for the test set were 0.0214, 0.126, 0.978, and 2.475, respectively. The results obtained show the superior prediction ability of the proposed model in the prediction of necroptosis inhibitory activities.

Bayesian Regularized Artificial Neural Network Modeling of the Anti‐protozoal Activities of 1‐Methylbenzimidazole Derivatives Against T. Vaginalis Infection

AbstractIn this work, we propose a quantitative structure‐activity relationship (QSAR) approach in order to model the anti‐protozoal activities of 24 derivatives of 1‐methylbenzimidazole against trichomoniasis vaginalis. A QSAR model was built using the Bayesian regularized artificial neural network (BR‐ANN). A total of 7 molecular descriptors, as the most feasible ones, were selected to build the QSAR model. Some of the selected descriptors seem to relate to the presence of ester group and chlorine atoms at the inner atmosphere of the compounds to increase the anti‐protozoal activity. The robustness and predictability of the proposed model were verified by the test data set and leave‐one‐out cross‐validation methods. The mean square errors (MSEs) and mean relative errors (MREs) for the test data set are 0.0049 and 0.82%, respectively, and for the leave‐one‐out method, they are 0.0031 and 0.75%, respectively. Finally, it was proved that the QSAR model is a useful tool in prediction of the anti‐protozoal activities of 1‐methylbenzimidazole compounds, and that it could be used to guide the synthesis of new 1‐methylbenzimidazole derivatives with better activities.

Recent Advances in Ligand-Based Drug Design: Relevance and Utility of the Conformationally Sampled Pharmacophore Approach

In the absence of three-dimensional (3D) structures of potential drug targets, ligand-based drug design is one of the popular approaches for drug discovery and lead optimization. 3D structure-activity relationships (3D QSAR) and pharmacophore modeling are the most important and widely used tools in ligand-based drug design that can provide crucial insights into the nature of the interactions between drug target and ligand molecule and provide predictive models suitable for lead compound optimization. This review article will briefly discuss the features and potential application of recent advances in ligand-based drug design, with emphasis on a detailed description of a novel 3D QSAR method based on the conformationally sample pharmacophore (CSP) approach (denoted CSP-SAR). In addition, data from a published study are used to compare the CSP-SAR approach to the Catalyst method, emphasizing the utility of the CSP approach for ligand-based model development.

QSPR predictions of heat of fusion of organic compounds using Bayesian regularized artificial neural networks

Computational approaches for the prediction of environmental pollutants' properties have great potential in rapid environmental risk assessment and management with reduced experimental cost. A quantitative structure–property relationship (QSPR) study was conducted to predict the heat of fusion of a set of organic compounds that have adverse effect on the environment. The forward selection (FS) strategy was used for descriptors selection. We examined the feasibility of using multiple linear regression (MLR), artificial neural networks (ANN) and Bayesian regularized artificial neural networks (BRANN) as linear and nonlinear methods. The QSPR models were validated by an external set of compounds that were not used in the model development stage. All models reliably predicted the heat of fusion of the organic compounds under study, whereas more accurate results were obtained by the BRANN model.

Genetic algorithm optimization in drug design QSAR: Bayesian-regularized genetic neural networks (BRGNN) and genetic algorithm-optimized support vectors machines (GA-SVM)

Many articles in "in silico" drug design implemented genetic algorithm (GA) for feature selection, model optimization, conformational search, or docking studies. Some of these articles described GA applications to quantitative structure-activity relationships (QSAR) modeling in combination with regression and/or classification techniques. We reviewed the implementation of GA in drug design QSAR and specifically its performance in the optimization of robust mathematical models such as Bayesian-regularized artificial neural networks (BRANNs) and support vector machines (SVMs) on different drug design problems. Modeled data sets encompassed ADMET and solubility properties, cancer target inhibitors, acetylcholinesterase inhibitors, HIV-1 protease inhibitors, ion-channel and calcium entry blockers, and antiprotozoan compounds as well as protein classes, functional, and conformational stability data. The GA-optimized predictors were often more accurate and robust than previous published models on the same data sets and explained more than 65% of data variances in validation experiments. In addition, feature selection over large pools of molecular descriptors provided insights into the structural and atomic properties ruling ligand-target interactions.

Simultaneous optimization by neuro-genetic approach of a multiresidue method for determination of pesticides in Passiflora alata infuses using headspace solid phase microextraction and gas chromatography

A simultaneous optimization strategy based on neuro-genetic approach has been applied to a HS-SPME-GC-ECD (Headspace Solid Phase Microextraction coupled to Gas Chromatography with Electron Capture Detection) method for simultaneous determination of the pesticides chlorotalonil, methyl parathion, malathion, α-endosulfan and β-endosulfan in herbal infusions of Passiflora alata (Dryander). Two types of extractive fibers were used: a home-made device coated by sol–gel process with polydimethylsiloxane–poly(vinyl alcohol) (PDMS/PVA) and a commercial PDMS. The effects of extraction parameters such as dilution of the infusion, extraction temperature and time, as well as sample ionic strength were evaluated through the Doehlert design. To find a model that could relate these extraction parameters with the extraction efficiency of all pesticide simultaneously, a Bayesian Regularized Artificial Neural Network (BRANN) approach was employed. Subsequently, Genetic Algorithm (GA) was applied to attain the optimum values from the model developed by the neural network. The use of the proposed approach allowed the determination of a single extraction condition that maximized the peak areas of all pesticides simultaneously, showing a promising and a suitable new procedure to the optimization process of complex analytical problems.

Linear and nonlinear modeling of antifungal activity of some heterocyclic ring derivatives using multiple linear regression and Bayesian-regularized neural networks

Antifungal activity was modeled for a set of 96 heterocyclic ring derivatives (2,5,6-trisubstituted benzoxazoles, 2,5-disubstituted benzimidazoles, 2-substituted benzothiazoles and 2-substituted oxazolo(4,5-b)pyridines) using multiple linear regression (MLR) and Bayesian-regularized artificial neural network (BRANN) techniques. Inhibitory activity against Candida albicans (log(1/C)) was correlated with 3D descriptors encoding the chemical structures of the heterocyclic compounds. Training and test sets were chosen by means of k-Means Clustering. The most appropriate variables for linear and nonlinear modeling were selected using a genetic algorithm (GA) approach. In addition to the MLR equation (MLR-GA), two nonlinear models were built, model BRANN employing the linear variable subset and an optimum model BRANN-GA obtained by a hybrid method that combined BRANN and GA approaches (BRANN-GA). The linear model fit the training set (n = 80) with r2 = 0.746, while BRANN and BRANN-GA gave higher values of r2 = 0.889 and r2 = 0.937, respectively. Beyond the improvement of training set fitting, the BRANN-GA model was superior to the others by being able to describe 87% of test set (n = 16) variance in comparison with 78 and 81% the MLR-GA and BRANN models, respectively. Our quantitative structure-activity relationship study suggests that the distributions of atomic mass, volume and polarizability have relevant relationships with the antifungal potency of the compounds studied. Furthermore, the ability of the six variables selected nonlinearly to differentiate the data was demonstrated when the total data set was well distributed in a Kohonen self-organizing neural network (KNN).

Comparison of Linear and Nonlinear Classification Algorithms for the Prediction of Drug and Chemical Metabolism by Human UDP‐Glucuronosyltransferase Isoforms.

Partial least squares discriminant analysis (PLSDA), Bayesian regularized artificial neural network (BRANN), and support vector machine (SVM) methodologies were compared by their ability to classify substrates and nonsubstrates of 12 isoforms of human UDP-glucuronosyltransferase (UGT), an enzyme “superfamily” involved in the metabolism of drugs, nondrug xenobiotics, and endogenous compounds. Simple two-dimensional descriptors were used to capture chemical information. For each data set, 70% of the data were used for training, and the remainder were used to assess the generalization performance. In general, the SVM methodology was able to produce models with the best predictive performance, followed by BRANN and then PLSDA. However, a small number of data sets showed either equivalent or better predictability using PLSDA, which may indicate relatively linear relationships in these data sets. All SVM models showed predictive ability (>60% of test set predicted correctly) and five out of the 12 test sets sho...

A multivariable approach for mapping sub-pixel land cover distributions using MISR and MODIS: Application in the Brazilian Amazon region

Accurate mapping of land cover at continental to global scales is currently limited by our ability to exploit the spatial, temporal, and radiometric characteristics of the available satellite data. Many ecologically and biogeochemically important landscape features are spatially extensive, but occur at scales much smaller than the ∼1-km footprint of wide-swath, polar orbiting radiometers. This is especially true for land cover changes associated with human activities. Satellite instruments that offer the appropriate spatial detail have much smaller swaths and longer repeat times, resulting in compositing intervals that are too large to resolve the time scales of these changes. In addition, the cost and effort associated with acquisition and processing of high-resolution data for large areas is often prohibitive. Methods for taking advantage of information contained in multiple-scale observations by combining data from high-resolution and moderate resolution sensors are thus of great current interest. In this paper, we retrieve land cover distributions in two different parts of the Brazilian Amazon region by estimating relationships between land cover fractions derived from 30-m resolution ETM+ and reflectance data from ∼1-km resolution MODIS and MISR. The scaling relationships are derived using a Bayesian-regularized artificial neural network (ANN) and compared to results using linear unmixing (LU). We explore the simultaneous use of two significant independent variables in terrestrial optical remote sensing, wavelength, and sun-sensor geometry, by combining nadir-adjusted MODIS reflectances in seven bands (VIS-SWIR) with multiangular (−71° to +71°) bidirectional reflectance data from MISR. This research was motivated by evidence from modeling and field studies demonstrating that: (a) the angular dependence of reflectance (e.g., from MISR) contains information about the structural composition of canopies that is complementary to the wavelength dependence; and (b) the SWIR portion of the spectrum (e.g., from MODIS) is sensitive to canopy moisture and shading conditions and, therefore, to the successional status of the ecosystem. This case study, using the Bayesian artificial neural network with combined MODIS-MISR data to estimate sub-pixel land cover fractions, yielded a quantitative improvement over spectral linear unmixing of single-angle, multispectral data. Our results suggest potential for broad-scale applicability despite a number of challenges related to tropical atmospheric conditions.

Use of automatic relevance determination in QSAR studies using Bayesian neural networks.

We describe the use of Bayesian regularized artificial neural networks (BRANNs) coupled with automatic relevance determination (ARD) in the development of quantitative structure-activity relationship (QSAR) models. These BRANN-ARD networks have the potential to solve a number of problems which arise in QSAR modeling such as the following: choice of model; robustness of model; choice of validation set; size of validation effort; and optimization of network architecture. The ARD method ensures that irrelevant or highly correlated indices used in the modeling are neglected as well as showing which are the most important variables in modeling the activity data. The application of the methods to QSAR of compounds active at the benzodiazepine and muscarinic receptors as well as some toxicological data of the effect of substituted benzenes on Tetetrahymena pyriformis is illustrated.