Analysis of the internal representations developed by neural networks for structures applied to quantitative structure--activity relationship studies of benzodiazepines

An application of recursive cascade correlation (CC) neural networks to quantitative structure-activity relationship (QSAR) studies is presented, with emphasis on the study of the internal representations developed by the neural networks. Recursive CC is a neural network model recently proposed for the processing of structured data. It allows the direct handling of chemical compounds as labeled ordered directed graphs, and constitutes a novel approach to QSAR. The adopted representation of molecular structure captures, in a quite general and flexible way, significant topological aspects and chemical functionalities for each specific class of molecules showing a particular chemical reactivity or biological activity. A class of 1,4-benzodiazepin-2-ones is analyzed by the proposed approach. It compares favorably versus the traditional QSAR treatment based on equations. To show the ability of the model in capturing most of the structural features that account for the biological activity, the internal representations developed by the networks are analyzed by principal component analysis. This analysis shows that the networks are able to discover relevant structural features just on the basis of the association between the molecular morphology and the target property (affinity).     

人工神经网络预测煤炭成浆浓度的研究

考虑煤炭的多种理化特性建立了成浆浓度的神经网络预测模型，对其数据预处理方法、学习率和中间层节点数等进行了深入讨论。水分、挥发分、分析基碳、灰分和氧等五个因子对于煤炭成浆性的预测起到主导作用。五因子、七因子和八因子神经网络模型对煤炭成浆浓度的预测误差分别为：0.53%、0.50%和0.74%，而现有回归分析方程的误差为1.15%，故神经网络模型比回归分析方程有更好的预测能力，尤以七因子模型最佳。     

Tuning neural and fuzzy-neural networks for toxicity modeling

The need for general reliable models for predicting toxicity has led to the use of artificial intelligence. We applied neural and fuzzy-neural networks with the QSAR approach. We underline how the networks have to be tuned on the data sets generally involved in modeling toxicity. This study was conducted on 562 organic compounds in order to establish models for predictive the acute toxicity in fish.     

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。     

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.     

The Learned Symmetry Concept in Revealing Quantitative Structure-Activity Relationships With Artificial Neural Networks

Abstract

A method to build QSAR models based on substituent constants for congeneric sets of compounds having several topologically equivalent substituent positions was proposed. The approach is based on the application of artificial neural networks (learning to construct nonlinear structure-activity relationships taking into account necessary symmetry properties of training set structures) to a training set expanded by adding the copies of compounds with the same activity values but with permuted assignment of equivalent substituent positions. The better predictive power of these constructed models, as compared with the performances of neural network models for non-expanded sets was demonstrated for the calcium channel blockers of 1,4-dihydropyridine type and for hallucinogenic phenylalkylamines.     

Validation of counter propagation neural network models for predictive toxicology according to the OECD principles: a case study

The OECD has proposed five principles for validation of QSAR models used for regulatory purposes. Here we present a case study investigating how these principles can be applied to models based on Kohonen and counter propagation neural networks. The study is based on a counter propagation network model that has been built using toxicity data in fish fathead minnow for 541 compounds. The study demonstrates that most, if not all, of the OECD criteria may be met when modeling using this neural network approach.     

Predicting Absorption-Distribution Properties of Neuroprotective Phosphine-Borane Compounds Using In Silico Modeling and Machine Learning

Phosphine-borane complexes are novel chemical entities with preclinical efficacy in neuronal and ophthalmic disease models. In vitro and in vivo studies showed that the metabolites of these compounds are capable of cleaving disulfide bonds implicated in the downstream effects of axonal injury. A difficulty in using standard in silico methods for studying these drugs is that most computational tools are not designed for borane-containing compounds. Using in silico and machine learning methodologies, the absorption-distribution properties of these unique compounds were assessed. Features examined with in silico methods included cellular permeability, octanol-water partition coefficient, blood-brain barrier permeability, oral absorption and serum protein binding. The resultant neural networks demonstrated an appropriate level of accuracy and were comparable to existing in silico methodologies. Specifically, they were able to reliably predict pharmacokinetic features of known boron-containing compounds. These methods predicted that phosphine-borane compounds and their metabolites meet the necessary pharmacokinetic features for orally active drug candidates. This study showed that the combination of standard in silico predictive and machine learning models with neural networks is effective in predicting pharmacokinetic features of novel boron-containing compounds as neuroprotective drugs.     

Assessment of applicability domain for multivariate counter-propagation artificial neural network predictive models by minimum Euclidean distance space analysis: A case study

Alongside the validation, the concept of applicability domain (AD) is probably one of the most important aspects which determine the quality as well as reliability of the established quantitative structure–activity relationship (QSAR) models. To date, a variety of approaches for AD estimation have been devised which can be applied to particular type of QSAR models and their practical utilization is extensively elaborated in the literature. The present study introduces a novel, simple, and effective distance-based method for estimation of the AD in case of developed and validated predictive counter-propagation artificial neural network (CP ANN) models through a proficient exploitation of the Euclidean distance (ED) metric in the structure-representation vector space. The performance of the method was evaluated and explained in a case study by using a pre-built and validated CP ANN model for prediction of the transport activity of the transmembrane protein bilitranslocase for a diverse set of compounds. The method was tested on two more datasets in order to confirm its performance for evaluation of the applicability domain in CP ANN models. The chemical compounds determined as potential outliers, i.e., outside of the CP ANN model AD, were confirmed in a comparative AD assessment by using the leverage approach. Moreover, the method offers a graphical depiction of the AD for fast and simple determination of the extreme points.     

Nondestructive determination of compound amoxicillin powder by NIR spectroscopy with the aid of chemometrics

Near-infrared (NIR) spectroscopy, in combination with chemometrics, enables nondestructive analysis of solid samples without time-consuming sample preparation methods. A new method for the nondestructive determination of compound amoxicillin powder drug via NIR spectroscopy combined with an improved neural network model based on principal component analysis (PCA) and radial basis function (RBF) neural networks is investigated. The PCA technique is applied to extraction relevant features from lots of spectra data in order to reduce the input variables of the RBF neural networks. Various optimum principal component analysis-radial basis function (PCA-RBF) network models based on conventional spectra and preprocessing spectra (standard normal variate (SNV) and multiplicative scatter correction (MSC)) have been established and compared. Principal component regression (PCR) and partial least squares (PLS) multivariate calibrations are also used, which are compared with PCA-RBF neural networks. Experiment results show that the proposed PCA-RBF method is more efficient than PCR and PLS multivariate calibrations. And the PCA-RBF approach with SNV preprocessing spectra is found to provide the best performance.     

辨识药物定量构效关系的模糊神经网络方法研究

提出一种基于遗传算法的新型模糊神经网络方法,用于计算Benzodiazepines(BZs)类药物的定量构效关系.这类模糊神经网络综合了神经网络、遗传算法与模糊逻辑的各自优势,具有优良的定量构效关系辨识能力,其学习速度较快,不易陷入局部最小区域;网络知识以模糊语言变量的形式加以表达,不仅易于理解,而且能有效地利用已有的专家经验.一旦通过学习获得规律后,不仅能很好地预测化合物的活性,还能对后续的药物分子设计提供有益的理论指导.     

Predicting physical-chemical properties of compounds from molecular structures by recursive neural networks

In this paper, we report on the potential of a recently developed neural network for structures applied to the prediction of physical chemical properties of compounds. The proposed recursive neural network (RecNN) model is able to directly take as input a structured representation of the molecule and to model a direct and adaptive relationship between the molecular structure and target property. Therefore, it combines in a learning system the flexibility and general advantages of a neural network model with the representational power of a structured domain. As a result, a completely new approach to quantitative structure-activity relationship/quantitative structure-property relationship (QSPR/QSAR) analysis is obtained. An original representation of the molecular structures has been developed accounting for both the occurrence of specific atoms/groups and the topological relationships among them. Gibbs free energy of solvation in water, Delta(solv)G degrees , has been chosen as a benchmark for the model. The different approaches proposed in the literature for the prediction of this property have been reconsidered from a general perspective. The advantages of RecNN as a suitable tool for the automatization of fundamental parts of the QSPR/QSAR analysis have been highlighted. The RecNN model has been applied to the analysis of the Delta(solv)G degrees in water of 138 monofunctional acyclic organic compounds and tested on an external data set of 33 compounds. As a result of the statistical analysis, we obtained, for the predictive accuracy estimated on the test set, correlation coefficient R = 0.9985, standard deviation S = 0.68 kJ mol(-1), and mean absolute error MAE = 0.46 kJ mol(-1). The inherent ability of RecNN to abstract chemical knowledge through the adaptive learning process has been investigated by principal components analysis of the internal representations computed by the network. It has been found that the model recognizes the chemical compounds on the basis of a nontrivial combination of their chemical structure and target property.     

Modelling and prediction of retention in high-performance liquid chromatography by using neural networks

Summary Multi-layer feed-forward neural networks trained with an error back-propagation algorithm have been used to model retention behaviour of liquid chromatography as a function of the composition of the mobile phases. Conventional hydro-organic and micellar mobile phases were considered. Accurate retention modelling and prediction have been achieved using mobile phases defined by two, three and four parameters. With micellar mobile phases, the parameters involved included the concentrations of surfactant and organic modifier, pH and temperature. It is shown that neural networks provide a competitive tool to model varied inherent nonlinear relationships of retention behaviour with respect to the mobile phase parameters. The soft models defined by the weights of the networks are capable of accommodating all types of linear and nonlinear relationships, neural networks being specially useful when the relationships between retention behaviour and the mobile phase parameters are unknown. However, to train neural networks more experimental points than with hard-modelling methods are required, hence the use of the networks is recommended only for those cases where adequate theoretical or empirical models do not exist.     

Application of Artificial Neural Networks for Analysis of 13C NMR Spectra of Fucoidans

A new approach for structure determination of native and O-desulfated fucoidans by the analysis of their ¹³C NMR spectra by artificial neural networks (ANNs) is described. Two ANN models were studied: the simple three-layer feed-forward network, which employs supervised learning, and the adaptive resonance theory (ART) network with unsupervised learning. Training sets for the networks were constructed using chemical shifts of synthetic oligofucosides. The results obtained demonstrate that both models worked better in the case of desulfated fucoidans, while the ART-type networks gave better results in sulfated (native) fucoidan structure elucidation.     

人工神经网络研究硝基苯化合物结构与毒性的关系

基于简单的化学基团描述符,应用人工神经网络研究了硝基苯类化合物对黑呆头鱼的毒性构效关系,并与多元线性回归相比较,结果显示了人工神经网络处理非线性问题的优越性.