雷蕾 王新洲 张黎 杨策 亢力 朱永亮 钱向平 叶祖光

摘要：目的 探讨计算机毒性预测技术和方法，为中药安全性评价提供新的途径。方法 采用Mold2软件（version 2.0.0）对7409个中药化学成分进行分子描述符计算，初步筛选后采用随机森林算法进行定量构效关系（QSAR）模型构建，筛选出最优预测模型，并对2010年版《中华人民共和国药典》所载83个有毒中药中有化学成分单体结构报道的60种中药（1692个化学成分）进行急性毒性的预测。结果 经筛选，最终得到数据7409条。当描述符集合为52个时，随机森林算法建模模型准确度和Kappa值较高，分别为0.712和0.436；根据最优分子描述符集（52个）表征的化学空间距离将化合物聚类分为3类，第1类化合物最优模型的准确度和Kappa值分别为0.666和0.476，第2类化合物最优模型的准确度和Kappa值分别为0.804和0.381，第3类化合物最优模型的准确度和Kappa值分别为0.709和0.373。预测60种中药（1692个化学成分）中含剧毒化合物0个、高毒化合物2个、中毒化合物172个和低毒化合物1518个。结论 采用QSAR模型对中药化学成分进行急性毒性预测研究可为联合用药和进一步实验研究提供参考。

关键词：定量构效关系；预测模型；急性毒性；中药化学成分；大鼠

DOI：10.3969/j.issn.1005-5304.2016.01.010

中图分类号：R2-05；R285.5 文献标识码：A 文章编号：1005-5304（2016）01-0043-04

QSAR Study on Toxicity of Chemical Components of Chinese Materia Medica and Acute Toxicity of Rats LEI Lei1， WANG Xin-zhou2， ZHANG Li1， YANG Ce1， KANG Li1， ZHU Yong-liang2， QIAN Xiang-ping2， YE Zu-guang3 （1. Institute of Information on TCM， China Academy of Chinese Medical Sciences， Beijing 100700， China； 2. Suzhou Neupharma Co. Ltd.， Suzhou 215123， China； 3. Institute of Chinese Materia Medica， China Academy of Chinese Medical Sciences， Beijing 100700， China）

Abstract： Objective To study computer toxicity prediction technology and predict the acute toxicity of Chinese materia medica； To provide a new way and method for safety evaluation of traditional Chinese medicine. Methods First， Mold2 software （version 2.0.0） was used to calculate molecular descriptors of 7409 chemical components. After preliminary screening of molecular descriptors， quantitative structure-activity relationship （QSAR） models were built up with Random Forest （RF） for screening the optimum prediction model. From the 83 kinds of toxic Chinese materia medica in Chinese Pharmacopoeia （2010 edition）， acute toxicity of 60 kinds of Chinese materia medica reported from monomer structure （1692 chemical components） were under prediction. Results Totally 7409 pieces of data were obtained. When the descriptors were 52， RF modeling accuracy and Kappa were the highest， 0.712 and 0.436 respectively. Compound clusters were divided into 3 types according to optimum molecule descriptors （52）. The accuracy and Kappa of the optimum model for the first type of compounds were 0.666 and 0.476 respectively； the accuracy and Kappa of the optimum model for the second type of compounds were 0.804 and 0.381 respectively； the accuracy and Kappa of the optimum model for the third type of compounds were 0.709 and 0.373 respectively. It was predicted that 60 kinds of Chinese materia medica containing 0 violent toxic compound， 2 high toxic compounds， 172 medium toxic compounds and 1518 low toxic compound. Conclusion QSAR model for prediction study on acute toxicity of chemical components of Chinese mareria medica can provide references combination medication and experimental studies.

Key words： quantitative structure-activity relationship； prediction model； acute toxicity； chemical components of Chinese materia medica； rats

急性毒性是指机体（人或动物）一次（或24 h内多次）接触外来化合物之后所引起的中毒效应。一般而言，进行药物毒理学研究时常首先进行急性毒性实验，从而对药物毒性有初步了解，为下一步深入毒理学研究提供基础数据。但是，急性毒性实……