论文部分内容阅读
本文应用网络科学的理论与方法,将组成分子的原子看作是网络节点,而一个分子的平面结构图就是由这些节点通过边连接而成,从而构造出分子结构网络,进一步计算出分子结构网络的平均度、平均路程长度等18个网络静态特征变量,作为构效分析的候选自变量。为了更好的进行构效分析,利用BP(Back Propagate,反向传播)人工神经网络从18个候选自变量中筛选出对活性贡献大的接近度中心性最大值等4个自变量,分别用支持向量机回归(SVR)和BP神经网络对分子结构网络的4个自变量与大鼠经口毒性LD_(50)进行定量构效关系(QSAR)研究,实验结果表明:基于分子结构网络静态参数的支持向量机回归模型具有良好的预测能力。说明应用分子结构网络静态参数建模,对具有同一(或多种)属性/活性的一类物质进行构效关系分析研究的这种新方法具有一定的应用前景。
In this paper, we use the theory and method of network science to regard the atoms of the constituent molecules as the network nodes, and the planar structure diagram of a molecule is formed by the connection of these nodes through the edges to construct the molecular structure network and further calculate the molecular structure network 18 network static characteristic variables, such as average degree and average length of route, were selected as candidate variables for structure-activity analysis. In order to make a better structure-activity analysis, we use BP (Back Propagate) artificial neural network to select four independent variables, including the maximal centrality of the contribution to the activity, from the 18 candidate variables. Support Vector Machine Regression (SVR) and BP Neural Network (QNAR) were used to study the quantitative structure-activity relationship (QSAR) of 4 independent variables of molecular structure network and rat oral LD_ (50). The experimental results showed that: based on the molecular structure network static parameters The support vector machine regression model has good predictive ability. It shows that this new method, which uses the static parameter modeling of molecular structure network, to study the structure-activity relationship of a group of substances with the same property (or properties) has certain application prospects.