Combining unsupervised and supervised artificial neural networks to predictaquatic toxicity

Most quantitative structure-activity relationship (QSAR) models are linear relationships and significant for only a limited domain of compounds. Here we propose a data-driven approach with a flexible combination of unsupervised and supervised neural networks able to predict the toxicity of a large set of different chemicals while still respecting the QSAR postulates. Since QSAR is applicable only to similar compounds, which have similar biological and physicochemical properties, large numbers of compounds are clustered before building local models, and local models are ensembled to obtain the final result. The approach has been used to develop models to predict the fish toxicity of Pimephales promelas and Tetrahymena pyriformis, a protozoan.     

A new optimization technique for artificial neural networks applied to prediction of force constants of large molecules

An artificial neural network (ANN) method for the prediction of force constants of chemical bonds in large, polyatomic molecules was developed. The force constant information evaluated is to be used for generating accurate estimates of the Hessian used in Newton-Raphson-type ab initio molecular structure optimization schemes. Different network topologies as well as a training procedure based on simulated annealing are evaluated. The results show that an ANN can be designed and trained to provide force constant information within a 1.5 to 5% error band even if the range of the force constants evaluated is very large (from triple bonds to hydrogen bridges). © 1995 by John Wiley & Sons, Inc.     

Esters flash point prediction using artificial neural networks

In this article, an artificial neural network to predict the flash point of 95 esters was implemented. Four variables were used for its development. A neural network with 4‐5‐8‐5‐1 topology was encountered to gain the best agreement of the experimental results with those predicted (square correlation coefficient (R²) and root mean square error were 0.99 and 5.46 K for the training phase and 0.96 and 13.02 K for the testing set). © 2012 Wiley Periodicals, Inc.     

Simultaneous kinetic determination of sulfite and sulfide using artificial neural networks

Artificial neural networks (ANNs) are proposed for the determination of sulfite and sulfide simultaneously. The method is based on the reaction between Brilliant Green (BG) as a colored reagent and sulfite and/or sulfide in buffered solution (pH 7.0) and monitoring the changes of absorbance at maximum wavelength of 628 nm. Experimental conditions such as pH, reagents concentrations, and temperature were optimized and training the network was performed using principal components (PCs) of the original data. The network architecture (number of input, hidden and output nodes), and some parameters such as learning rate (η) and momentum (α) were also optimized for getting satisfactory results with minimum errors. The measuring range was 0.05-3.6 μg ml⁻¹ for both analytes. The proposed method has been successfully applied to the quantification of the sulfite and sulfide in different water samples.     

Accurate prediction of rate constants of Diels-Alder reactions and application to design of Diels-Alder ligation

Bioorthogonal reactions are useful tools to gain insights into the structure, dynamics, and function of biomolecules in the field of chemical biology. Recently, the Diels-Alder reaction has become a promising and attractive procedure for ligation in bioorthogonal chemistry because of its higher rate and selectivity in water. However, a drawback of the previous Diels-Alder ligation is that the widely used maleimide moiety as a typical Michael acceptor can readily undergo Michael addition with nucleophiles in living systems. Thus, it is important to develop a nucleophile-tolerant Diels-Alder system in order to extend the scope of the application of Diels-Alder ligation. To solve this problem, we found that the theoretical protocol M06-2X/6-31+G(d)//B3LYP/6-31G(d) can accurately predict the activation free energies of Diels-Alder reactions with a precision of 1.4 kcal mol(-1) by benchmarking the calculations against the 72 available experimental data. Subsequently, the electronic effect and ring-strain effect on the Diels-Alder reaction were studied to guide the design of the new dienophiles. The criteria of the design is that the designed Diels-Alder reaction should have a lower barrier than the Michael addition, while at the same time it should show a similar (or even higher) reactivity as compared to the maleimide-involving Diels-Alder ligation. Among the designed dienophiles, three substituted cyclopropenes (i.e. 1,2-bis(trifluoromethyl)-, 1,2-bis(hydroxylmethyl)- and 1,2-bis(hydroxylmethyl)-3-carboxylcyclopropenes) meet our requirements. These substituted cyclopropene analogs could be synthesized and they are thermodynamically stable. As a result, we propose that 1,2-bis(trifluoromethyl)-, 1,2-bis(hydroxylmethyl)- and 1,2-bis(hydroxylmethyl)-3-carboxylcyclopropenes may be potential candidates for efficient and selective Diels-Alder ligation in living systems.     

Using artificial neural networks for prediction of organic acid partition coefficients

Aqueous/organic phase partition coefficients of organic acids were predicted using an artificial neural network (ANN) algorithm taking benzoic acid derivatives as examples. The partition coefficients were determined by extraction of the acids from aqueous salt solutions with hydrophilic solvents (BuⁿOH, BuⁱOH, and Bu^tOH). Using the ANN approach makes it possible to obtain quantitative information on the values of the title parameters. Published in Russian in Izvestiya Akademii Nauk. Seriya Khimicheskaya, No. 2, pp. 207—212, February, 2006.     

Optimization by experimental design and artificial neural networks of the ion-interaction reversed-phase liquid chromatographic separation of twenty cosmetic preservatives

Particular attention are recently receiving antimicrobial agents added as preservatives in hygiene and cosmetics commercial products, since some of them are suspected to be harmful to the human health. The preservatives used belong to different classes of chemical species and are generally used in their mixtures. Multi-component methods able to simultaneously determinate species with different chemical structure are therefore highly required in quality control analysis. This paper presents an ion interaction RP-HPLC method for the simultaneous separation of the 20 typical antimicrobial agents most used in cosmetics and hygiene products, that are: benzoic acid, salicylic acid, 4-hydroxybenzoic acid, methyl-, ethyl-, propyl-, butyl-, benzyl-benzoate, methyl-, ethyl-, propyl-, butyl-, benzyl-paraben, o-phenyl-phenol, 4-chloro-m-cresol, triclocarban, dehydroacetic acid, bronopol, sodium pyrithione and chlorhexidine. For the development of the method and the optimization of the chromatographic conditions, an experimental design was planned and models were built by the use of artificial neural network to correlate the retention time of each analyte to the variables and their interactions. The neuronal models developed showed good predictive ability and were used, by a grid search algorithm, to optimize the chromatographic conditions for the separation of the mixture.     

A priori rate constants for kinetic modeling

The advent of computer-aided methods for constructing detailed kinetic models of multicomponent reacting systems provides fresh motivation for the development of efficient and accurate methods for estimating rate constants. There is now the real likelihood that a priori rate estimates, formerly of primarily academic interest, could directly impact major public policy and business decisions. This opportunity brings many challenges. The process of building a computer model for a real-world system can require hundreds of thousands of rate estimates, making most existing rate calculation techniques impractical. Also, the demands for tight error bars on model predictions used to make major decisions often imply levels of accuracy unachievable with existing rate calculation techniques. Past and recent progress towards developing fast and accurate rate estimates is selectively reviewed, and our methodology is outlined. New rate estimates for several types of reactions involving O and HO₂ are presented. Several technical issues requiring further work by the theoretical chemistry community are highlighted. Electronic supplementary material to this paper can be obtained by using the Springer Link server located at http://dx.doi.org/10.1007/s00214-002-0368-4. Received: 6 February 2002 / Accepted: 2 June 2002 / Published online: 2 October 2002 Acknowledgements. This work was partially supported by the National Computational Science Alliance under Grant CTS010006 N and utilized the Origin 2000 High-performance Computing and UniTree Mass Storage systems. We are grateful for financial support from the EPA Center for Airborne Organics, the NSF CAREER program, Alstom Power, Dow Chemical, and the Office of Basic Energy Science, U.S. Department of Energy through grant DE-FG02-98ER14914. The authors acknowledge the contribution of Hans-Heinrich Carstensen in the initial stages of this work. Correspondence to: W.H. Green Jr. e-mail: whgreen@mit.edu Electronic supplementary material to this paper can be obtained by using the Springer LINK server located at http://dx.doi.org/10.1007/s00214-002-0368-4     

Modeling of the relationship between electroosmotic flow and separation parameters in capillary zone electrophoresis using artificial neural networks and experimental design

The prediction of migration time of electroosmotic flow (EOF) marker was achieved by applying artificial neural networks (ANN) model based on principal component analysis (PCA) and standard normal distribution simulation to the input variables. The voltage of performance, the temperature in the capillary, the pH and the ionic strength of background electrolytes (BGE) were applied as the input variables to ANN. The range of the performance voltage studied was from 15 to 27 kV, and that of the temperature in the capillary was from 20 to 30 °C. For the pH values studied, the range was from 5.15 to 8.04. The range of the ionic strength investigated in this paper was from 0.040 to 0.097. The prediction abilities of ANN with different pre-processing procedure to the input variables were compared. Under the same performance conditions, the average prediction error of the migration time of the EOF marker was 5.46% with RSD = 1.76% according to 10 parallel runs of the optimized ANN structure by the proposed approach, and that of the 10 parallel predictions of the optimal ANN structure for the different performance conditions was 12.95% with RSD = 2.29% according to the proposed approach. The study showed that the proposed method could give better predicted results than other approaches discussed.     

Quantitative chiral analysis in capillary electrophoresis from unresolved peaks using derivative electropherograms,experimental design,and artificial neural networks

Quantitative capillary electrophoretic analysis of chiral compounds might be difficult or even impossible when baseline separation is not reached. In this work, the use of n-th derivative of the electropherogram was studied and examined on model and experimental data. The electropherograms should be first smoothed using Savitzky-Golay method and the quantitative analysis is then possible using either a graphical method or multivariate calibration applying a combination of experimental design (ED) and artificial neural networks (ANNs). The best results were obtained for the first derivative, higher derivatives are not suitable because of noise accumulation. The method was applied to real experimental data to quantify chiral amino acids from unresolved peaks, but it is applicable for quantitative analysis of any other chiral analytes from poorly resolved peaks. Precision of analysis from partially resolved peaks reached was about +/- 3.2% relative standard deviation.     

Optimization of the separation of a group of triazine herbicides by micellar capillary electrophoresis using experimental design and artificial neural networks

The micellar electrokinetic chromatography separation of a group of triazine compounds was optimized using a combination of experimental design (ED) and artificial neural network (ANN). Different variables affecting separation were selected and used as input in the ANN. A chromatographic exponential function (CEF) combining resolution and separation time was used as output to obtain optimal separation conditions. An optimized buffer (19.3 mM sodium borate, 15.4 mM disodium hydrogen phosphate, 28.4 mM SDS, pH 9.45, and 7.5% 1-propanol) provides the best separation with regard to resolution and separation time. Besides, an analysis of variance (ANOVA) approach of the MEKC separation, using the same variables, was developed, and the best capability of the combination of ED-ANN for the optimization of the analytical methodology was demonstrated by comparing the results obtained from both approaches. In order to validate the proposed method, the different analytical parameters as repeatability and day-to-day precision were calculated. Finally, the optimized method was applied to the determination of these compounds in spiked and nonspiked ground water samples.     

人工神经网络阻抑动力学光度法同时测定邻苯二酚和间苯二胺

在硫酸性介质中,Fe(Ⅲ)能够催化H2O2氧化中性红褪色反应,邻苯二酚和间苯二胺都能阻抑该催化氧化褪色反应的速度,研究发现：两者对Fe(Ⅲ)催化H2O2氧化中性红褪色反应阻抑作用不具有加和性,根据这一现象,用人工神经网络处理非线性体系的优势进行数据处理,从而建立了一种新的测定邻苯二酚和间苯二胺混合物的人工神经网络阻抑动力学光度法。对5组混合样品进行测定,回收率均在95%-105%之间。     

Application of generalized artificial neural networks coupled with an orthogonal design to optimization of a system for the kinetic spectrophotometric determination of Hg(II)

A simple and sensitive kinetic method for the determination of traces of mercury (70-760 ng ml⁻¹) based on its inhibitory effect on the addition reaction between methyl green and sulfite ion is proposed. The reaction was monitored spectrophotometrically by measuring the decrease in absorbance of methyl green at 596 nm between 2 and 4 min using a fixed time method. Artificial neural networks with back propagation algorithm coupled with an orthogonal array design were applied to the modeling of the proposed kinetic system and optimization of experimental conditions. An orthogonal design was utilized to design the experimental protocol, in which pH, concentration of sulfite, temperature, and concentration of methyl green were varied simultaneously. Optimum experimental conditions in term of sensitivity were generated by using ANNs. The rate of decrease in absorbance is inversely proportional to the concentration of Hg(II) over entire concentration range tested (100-550 ng ml⁻¹) with a detection limit of 45 ng ml⁻¹ and a relative standard deviation at 200-400 ng ml⁻¹ Hg(II) of 3.2% (n=5). A simple preconcentration step improved the limit of detection and linear dynamic range of the method to about 8 and 12-760 ng ml⁻¹, respectively, by about 10 times enrichment of mercury between 12 and 75 ng ml⁻¹. The method was based on enrichment of Hg(II) from dilute samples on an anionic ion exchanger fixed on a plastic strip and was applied to the determination of Hg(II) in environmental samples with satisfactory results.     

Application of artificial neural networks to predictions in flow-injection spectrophotometry

It is demonstrated that predictions can be obtained in spectrophotometric flow-injection analysis (FIA) based on an experimental parameter, that is, the degree of reaction, which takes into account the hydrodynamic and chemical characteristics of the spectrophotometric reaction used. The search algorithm is based on constructing a model of a chemical-analytical process using a learning artificial neural network that enables the prediction of the degree of reaction for some reagents not studied yet. The trained neural network is used for the a priori evaluation and comparison of a number of reagents for the determination of aluminum by FIA.     

The prediction of total anthocyanin concentration in red-grape homogenates using visible-near-infrared spectroscopy and artificial neural networks

This study compares the performance of partial least squares (PLS) regression analysis and artificial neural networks (ANN) for the prediction of total anthocyanin concentration in red-grape homogenates from their visible-near-infrared (Vis-NIR) spectra. The PLS prediction of anthocyanin concentrations for new-season samples from Vis-NIR spectra was characterised by regression non-linearity and prediction bias. In practice, this usually requires the inclusion of some samples from the new vintage to improve the prediction. The use of WinISI LOCAL partly alleviated these problems but still resulted in increased error at high and low extremes of the anthocyanin concentration range. Artificial neural networks regression was investigated as an alternative method to PLS, due to the inherent advantages of ANN for modelling non-linear systems. The method proposed here combines the advantages of the data reduction capabilities of PLS regression with the non-linear modelling capabilities of ANN. With the use of PLS scores as inputs for ANN regression, the model was shown to be quicker and easier to train than using raw full-spectrum data. The ANN calibration for prediction of new vintage grape data, using PLS scores as inputs, was more linear and accurate than global and LOCAL PLS models and appears to reduce the need for refreshing the calibration with new-season samples. ANN with PLS scores required fewer inputs and was less prone to overfitting than using PCA scores. A variation of the ANN method, using carefully selected spectral frequencies as inputs, resulted in prediction accuracy comparable to those using PLS scores but, as for PCA inputs, was also prone to overfitting with redundant wavelengths.