目的 为了寻找新一代紫杉醇类抗癌药物，研究紫杉醇类似物分子结构和抗癌活性（PID_act）之间的定量构效关系。方法 基于电性状态拓扑指数和电性距离矢量，应用最佳子集回归的方法建立了PID_act值和紫杉醇类似物分子结构的定量结构–活性相关（QSAR）模型，并对模型进行了交互检验和外部检验，用43个紫杉醇类似物训练集样本构建的QSAR模型预测了外部10个检验集的PID_act值。结果 所建立模型具有较高的估计相关系数及LOO（leave-one-out）检验相关系数。结论 模型具有良好估计能力与稳定性，训练集模型具有良好预测能力。

Objective To develop a new generation of anticancer drugs of paclitaxel, the quantitative structure-anticancer activity (PID_act) relationship (QSAR) of paclitaxel analogues was studied. Methods Based on the electrotopological state index and electronegativity distance vector, the QSAR model was developed by Leaps-Bounds regression (best subset, LBR), and was validated using cross and external-validation. PID_act values of ten external validation were predicted by the QSAR model built from the training set with 43 paclitaxel analogues. Results The model had higher calibrated correlation coefficient and LOO (leave-one-out) validated correlation coefficient. Conclusion The model has estimated ability and good stability, and the training set model has good predictive ability.

江苏省自然科学基金资助项目（09KJD150012）；徐州市科技计划研究项目（XM08C015；XX10A060）