改进的MC_GEP算法与嘌呤骨架类Hsp90抗癌活性的定量结构活性关系研究  

QSAR study on antitumor activity of 35 kinds of purine derivatives with Hsp90 based on an improved MC_GEP algorithm

作　　者：付伟忠[1] 张运陶[1] 

机构地区：[1]西华师范大学应用化学研究所,四川南充637002

出　　处：《计算机与应用化学》2010年第5期635-639,共5页Computers and Applied Chemistry

摘　　要：以35种嘌呤骨架类热休克蛋白90(Hsp90)化合物为研究对象,以文献[3]的8个变量构成自变量集,提出1种改进的MC GEP算法对该类化合物抗癌活性pEC_(50)做定量结构活性关系研究。按文献[3]对所有35种嘌呤类化合物建模,本文GEP模型与文献[3]的回归模型计算结果决定系数R^2分别为0.821 4和0.738 0。进一步用k均值聚类算法将此35种化合物划分为训练集和预测集,分别采用改进的MC GEP算法、v-SVM和ε-SVM算法基于训练集建模,本文建立的GEP模型训练和预测结果R^2分别为0.808 0和0.745 5,而v-SVM和ε-SVM模型对预测集的预测结果R^2分别为0.204 6和0.410 3,均低于0.5。研究表明,本文提出的改进MC GEP算法函数发现能力较强,建立的QSAR模型预测性能好。An improved MC_GEP algorithm is proposed for QSAR study on antitumor activity of 35 kinds ofpurine derivatives with heat shock protein 90 （Hsp90） compounds, taking the eight variables from the Reference [3] as independent variables andpEC5o as the dependent variable. According to the Reference [3], we have established QSAR models with all 35 kinds of purine derivatives with Hsp90. In this paper, the determination coefficient R2 of GEP model results and the regression model of the Reference [3] were 0.821 4 and 0.738 0 respectively. For further study, 35 kinds of compounds are divided into training set and prediction set, taking the improved MC GEP, v-SVM and the ε-SVM algorithm are based on the training set for modeling. The R2 of GEP model in training and predictionset are 0.808 0 and 0.745 5 respectively. The R2 ofv-SVM and the e-SVM model predicting results are 0.204 6 and 0.410 3 respectively, less than 0.5. The results show that the improved MC_GEP algorithm has good ability of function finding and the predict performance of QSAR models is good.

关 键 词：热休克蛋白90 定量结构活性关系 改进MC_GEP 线性解码 

分 类 号：O641.12[理学—物理化学]