基于电性距离矢量研究多氯联苯生物降解速率常数

通过多元线性回归和人工神经网络方法建立66种多氯联苯生物降解速率常数(K1)的定量构效关系(QSAR). 基于电性距离矢量(Mk),建立了lnK1的最佳三参数(M91、M25和M15)线性模型,其传统相关系数(R2)、交叉验证系数(Rcv2)分别为0.833、0.809。经R2、Rcv2、VIF、FIT、AIC检验,所建模型具有较强的稳定性和良好的预测能力. 将M91、M25、M15作为人工神经网络的输入层结点,采用3:10:1的网络结构,利用BP算法获得了一个令人满意的lnK1模型,训练集、验证集、测试集和总体的R2依次为0.991、0.995、0.997和 0.993。与多元线性回归模型相比,非线性lnK1-BP模型具有更好的预测能力。这两种回归方法相辅相成,线性回归方法为神经网络模型提供了具体的物理解释,而神经网络方法为线性模型提供了更准确的预测结果。     

计算烷烃沸点的新方法-基团键贡献法

根据分子中基团的特性和连接性,将基团贡献法和化学键贡献法结合在一起,发展了一种直接根据分子结构信息计算烷烃沸点的新方法-基团键贡献法,此方法同时具有基团贡献法和化学键贡献法的特点。对753种烷烃(C2～C100)的计算结果表明,沸点计算值与实验值的一致性令人满意,平均误差0.46%。     

聚酰亚胺玻璃化转变温度预测:基团贡献加和法与未知基团赋值

基团贡献加和法(GAP)假设聚合物性质来自于重复单元中各次级基团的贡献, 因此可以通过计算基团贡献值的加和值预测聚合物性质. van Krevelen建立了基团贡献加和法, 计算了数十种聚合物的性质, 包括常用的溶解度参数、 熔点和玻璃化转变温度(T_g)等参数. 聚酰亚胺是由二酐和二胺缩合反应得到的一类高性能聚合物, 其中T_g是决定聚酰亚胺使用温度范围的关键性质. 因此准确预测聚酰亚胺的T_g有助于优化和筛选单体分子结构. 本文首先利用van Krevelen提供的普适性基团贡献值计算了74种聚酰亚胺的T_g, 发现计算值与实验值具有较好的相关性(R²=0.88, s=21 K), 但存在系统误差, 如二者线性拟合斜率为0.78, 远偏离1. 由于普适性贡献值来自于不同聚合物的数据迭代, 对聚酰亚胺体系适用性较差, 必须对基团贡献值进行校正. 本文系统性地提高了刚性基团的贡献值, 同时降低了柔性基团的贡献值. 利用校正后的基团贡献值重新计算了T_g, 其与实验值具有更好的相关性(R²=0.88, s=18 K)和一致性(线性拟合斜率为0.94). 进一步使用上述校正后的已知基团贡献值对未知的7种二酐基团和6种二酐或二胺中的子基团进行赋值. 训练组(82个聚酰亚胺)和测试组(35个聚酰亚胺)数据验证了这13个基团贡献值的可靠性. 本文建立的基团贡献值校正方法和对未知基团的赋值法也可以推广应用于其它芳杂环类聚合物.     

基于三维荧光光谱的枸杞鉴别方法

以枸杞干果为研究对象,将三维荧光光谱技术与平行因子分析法、BP神经网络相结合,建立枸杞定性鉴别模型。采用固体样品支架测得枸杞粉末样品的三维荧光光谱,利用平行因子分析方法对预处理后的三维荧光矩阵进行三线性分解得到2个主因子的浓度得分,然后将浓度得分作为BP神经网络的输入向量,建立枸杞的人工神经网络鉴别模型。利用所建模型对待测样品进行预测,预测正确率为100%。结果表明,平行因子分析结合BP神经网络建立的枸杞产地鉴别模型,能够快速准确地鉴别宁夏枸杞。     

应用基团键贡献法计算烷烃和环烷烃的折光指数

根据分子中基团的特性和连接性,发展了一种直接根据分子结构信息计算烷烃和环烷烃折光指数的新方法－基团键贡献法,该方法既考虑分子中基团的特性,又考虑基团之间的连接性(化学键),同时具有基团贡献法和化学键贡献法的特点。应用基团键贡献法对950种烷烃和环烷烃折光指数的计算结果表明,计算值与实验值的一致性令人满意,平均误差0.11%,进一步对聚乙烯、聚丙烯和聚1-丁烯等聚合物的折光指数进行预测,也取得了令人满意的结果。     

预测烷烃密度的新方法: 基团键贡献法

根据分子结构的特点,通过用染色矩阵和邻接矩阵对分子结构进行矩阵化表征,发展了一种根据分子结构信息烷烃密度的新方法---基团键贡献法。该方法有机地将基团贡献法和化学键贡献法结合在一起,既考虑了分子中基团的特性,又考虑了基团之间的连接性(化学键),具有基团贡献法和化学键贡献法的特点。对658种烷烃的计算结果表明,密度预测值十分接近实验值,平均误差0.245%,进一步外推对聚乙烯、聚丙烯和聚1-丁烯等聚合物的密度进行预测,也取得了令人满意的结果。     

支持向量机分类和回归用于肽的QSAR研究

使用支持向量机技术对两类肽化合物体系进行了分类和回归研究,并将其系统地与K最邻近法、多元线性回归、偏最小二乘、人工神经网络进行了比较。结果表明,对于小样本、非线性问题,支持向量机具有较强的稳定性能及泛化能力,在大多数情况下能够得到优于传统方法的建模效果。对于分类问题,支持向量机对训练集和测试集都达到了100%的分类正确率;对于回归问题,支持向量机虽对训练集样本拟合效果略低于人工神经网络,但对外部测试集却表现出较强的预测能力。     

芳香胺类化合物N—H键离解能的定量构效关系研究

基于启发式方法(HM)和BP人工神经网络方法建立了5个参数的定量结构性质关系(QSPR)模型, 用于预测80个芳香胺类化合物N—H键的键离解能(BDE). 通过两种方法分别建立了线性和非线性的QSPR模型, 相关系数R分别为0.823和0.976. 通过对模型的稳定性和预测能力进行比较, 发现BP人工神经网络方法能够更好地预测芳香胺类化合物N—H键的BDE值.     

碳氢化合物燃烧中间体Lennard-Jones系数的计算

在已有的基团贡献法公式的基础上,提出了一种新的基团贡献法公式,并通过拟合250种化合物(包括185种稳定化合物临界性质的实验值和65种自由基临界性质的计算值)的临界性质得到了40种基团的贡献值,并用于预测未知化合物的临界性质.选取了训练集以外的、有临界性质实验值的30种化合物作为独立测试集,用于验证所建模型对临界性质的预测能力,T_C和P_C平均绝对偏差分别为8.52%和16.83%.结果表明,预测结果和实验值相吻合,该模型可以用于大分子化合物及自由基的临界性质预测.根据临界性质与Lennard-Jones(L-J)系数的经验关系式,预测了碳氢化合物燃烧中间体的L-J系数,得到独立测试集46种碳氢化合物的L-J系数,与文献值接近,T_C和P_C的平均绝对偏差分别为9.88%和9.96%.比较了训练集中烷烃自由基·C_6H_(13)、烯烃自由基·C_5H_9和炔烃自由基·C_5H_7同分异构体的L-J系数,同时,将己烷自由基·C_6H_(13)与相似的邻近烷烃C_6H_(14)的L-J系数进行比较,发现同分异构体之间或相似化合物之间L-J系数有较大偏差.此外,对缺少L-J系数的114种常见碳氢化合物自由基进行了预测.这对于碳氢化合物的燃烧模拟及基元反应中压强相关的速率常数计算有重要意义.     

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。     

Quantitative structure chromatography relationships in reversed-phase high performance liquid chromatography: Prediction of retention behaviour using theoretically derived molecular properties

Summary The use of theoretically calculated molecular properties as predictors for retention in reversed-phase HPLC has been explored. HPLC retention times have been measured for a series of 47 substituted aromatic molecules in three solvent mixtures and steric and electronic properties of these compounds have been derived using semi-empirical molecular orbital and empirical theoretical methods. A subset of the experimental data (a training set) was used to derive property-retention time relationships and the remaining data were then used to test the predictive capability of the methods.Good retention time prediction was possible using derived regression equations for individual solvents and after including solvent parameters it was possible to predict retention for all solvents using a single equation. This method showed that the most useful properties were calculated log P and the calculated dipole moment of the solutes, and the calculated solvent polarisability. In addition, 90% of the data were used to train an artificial neural network and the remaining 10% of the data used to test the network; excellent prediction was obtained, the neural network approach being as successful as the regression analysis.     

Neural network studies : Part 3. Prediction of partition coefficients

In a previous paper (N. Bodor, A. Harget and M.-J. Huang, J. Am. Chem. Soc., 113 (1991) 9480) we demonstrated the utility of a neural network approach in the estimation of the aqueous solubility of organic compounds. This approach has now been extended to the prediction of partition coefficients. A training set of AM1 calculated properties and experimental values for 302 compounds was used and, after training, the neural network was tested for its ability to predict the partition coefficients of 21 compounds not included in the training set. We also tested six more compounds with molecular properties out of the training set property range. A comparison was made with the results obtained from a previous study which had used a regression analysis approach (N. Bodor and M.-J. Huang, J. Pharm. Sci., 81 (1992) 272). The neural network results compared favorably with those given by the regression analysis approach, both for the training set and for the new compounds.     

Prediction of n-octanol/water partition coefficients from PHYSPROP database using artificial neural networks and E-state indices

A new method, ALOGPS v 2.0 (http://www.lnh.unil.ch/~itetko/logp/), for the assessment of n-octanol/water partition coefficient, log P, was developed on the basis of neural network ensemble analysis of 12 908 organic compounds available from PHYSPROP database of Syracuse Research Corporation. The atom and bond-type E-state indices as well as the number of hydrogen and non-hydrogen atoms were used to represent the molecular structures. A preliminary selection of indices was performed by multiple linear regression analysis, and 75 input parameters were chosen. Some of the parameters combined several atom-type or bond-type indices with similar physicochemical properties. The neural network ensemble training was performed by efficient partition algorithm developed by the authors. The ensemble contained 50 neural networks, and each neural network had 10 neurons in one hidden layer. The prediction ability of the developed approach was estimated using both leave-one-out (LOO) technique and training/test protocol. In case of interseries predictions, i.e., when molecules in the test and in the training subsets were selected by chance from the same set of compounds, both approaches provided similar results. ALOGPS performance was significantly better than the results obtained by other tested methods. For a subset of 12 777 molecules the LOO results, namely correlation coefficient r(2)= 0.95, root mean squared error, RMSE = 0.39, and an absolute mean error, MAE = 0.29, were calculated. For two cross-series predictions, i.e., when molecules in the training and in the test sets belong to different series of compounds, all analyzed methods performed less efficiently. The decrease in the performance could be explained by a different diversity of molecules in the training and in the test sets. However, even for such difficult cases the ALOGPS method provided better prediction ability than the other tested methods. We have shown that the diversity of the training sets rather than the design of the methods is the main factor determining their prediction ability for new data. A comparative performance of the methods as well as a dependence on the number of non-hydrogen atoms in a molecule is also presented.