论文部分内容阅读
从表征分子的66个结构参量和拓扑指数出发,经主成分分析(PCA)得到一种包含134个氨基酸体系的特征描述子:氨基酸拓扑与结构信息矢量(VTSA).将其应用于58个血管紧张素转化酶抑制剂(ACEI)和88个弹性蛋白酶模拟底物催化动力体系(ESCK)的定量构效关系(QSAR)或定量序效模拟(QSAM)研究中,结合遗传偏最小二乘(GPLS)、支持向量机回归(SVMR)以及本文设计的免疫神经网络(INN)技术,成功建立了上述两个肽类似物样本集定量预测模型,并取得优于已有文献报道的结果:ACEI,Rcu2≥0.82,Rcu2≥0.77,Ermse≤0.44(GPLS+SVM);ESCK,Rcu2≥0.84,Rcu2≥0.82,Ermse≤0.20(GPLS+INN).
Based on the 66 structural parameters and topological indices of molecular characterization, a characteristic descriptor of 134 amino acids was obtained by principal component analysis (PCA): amino acid topology and structure information vector (VTSA) (QSAR) or quantitative QSAM (quantitative structure-activity modeling) studies of ACE inhibitors and 88 elastase-catalyzed substrate kinetic systems (ESCK), combined with genetic partial least squares (GPLS ), Support vector machine regression (SVMR) and the immune neural network (INN) technology designed in this paper, we successfully established the quantitative prediction model of the two peptide analogs sample sets and achieved better results than those reported in the literature: ACEI, Rcu2 ≧ 0.82, Rcu2 ≧ 0.77, Ermse ≦ 0.44 (GPLS + SVM); ESCK, Rcu2 ≧ 0.84, Rcu2 ≧ 0.82, Ermse ≦ 0.20 (GPLS + INN).