癌变·畸变·突变 ›› 2015, Vol. 27 ›› Issue (5): 399-403.doi: 10.3969/j.issn.1004-616x.2015.05.015

• 技术与方法 • 上一篇    下一篇

应用PLS-QSAR模型预测纳米金属氧化物的毒性

邓景景, 王萌, 唐萌, 刘冉, 闵捷, 余小金   

  1. 东南大学公共卫生学院, 江苏 南京 210009
  • 收稿日期:2015-04-28 修回日期:2015-07-25 出版日期:2015-09-30 发布日期:2015-09-30
  • 通讯作者: 余小金,E-mail:xiaojinyu@seu.edu.cn E-mail:xiaojinyu@seu.edu.cn
  • 作者简介:邓景景,E-mail:djj3554455@163.com。
  • 基金资助:

    国家重大科学研究计划项目(2011CB933404)

The application of PLS-QSAR model to predict the toxicity of nanometer metal oxide

DENG Jingjing, WANG Meng, TANG Meng, LIU Ran, MIN Jie, YU Xiaojin   

  1. Department of Preventive Medicine, School of Public Health, Nanjing 210009, Jiangsu, China
  • Received:2015-04-28 Revised:2015-07-25 Online:2015-09-30 Published:2015-09-30

摘要:

目的:探讨基于偏最小二乘法(PLS)的定量构效关系(QSAR)模型(即PLS-QSAR模型)在纳米金属氧化物细菌毒性预测中的应用策略,为同类毒理学研究提供新的方法学选择。方法:应用PLS-QSAR模型分析纳米金属氧化物的结构描述符与细菌毒性之间的关系,分别采用3种主成分确定方法和4种变量选择方法建立模型,并对所建模型进行内部验证和外部验证,评价模型的预测能力和可解释性,以确定最优模型。结果:结合变量重要性指标(VIP)和交叉验证指标的向前变量选择法(FPVQ)和基于检验的主成分确定方法(CVtest)得到的模型,拟合能力(R2=0.974)、稳定性(QCV=0.951)、预测能力(QEXT=0.775)方面均较好。结论:PLS-QSAR模型可作为一个重要的方法用于纳米金属氧化物的毒性预测,对于变量选择和主成分确定的策略需要结合数据特征具体探讨。

关键词: 偏最小二乘法, 定量构效关系, 纳米金属氧化物, 毒性

Abstract:

OBJECTIVE: To explore the application of partial least squares-quantities structure activity relationship (PLS-QSAR) model to predict the bacterial toxicity of nanometer metal oxide, in order to provide methodological support for toxicology research. METHODS:We built 12 models through 3 ways to determine principal components and 4 ways of variables selection from published data of nanometer metal oxide (EC50) and chose the best model based on internal and external validations. RESULTS:We chose the best model obtained using forward selection of variables combining variable importance in projection(VIP) and cross validation, and CVtest principle components determination. The model had the better fitness(R2=0.974), stability (QCV=0.951) and prediction (QEXT=0.775). CONCLUSION:PLS-QSAR can be applied to predict the bacterial toxicity of nanometer metal oxide as an alternative method and it may be worthwhile to explore variable selections and determine number of principle component.

Key words: partial least squares, quantities structure activity relationship, nanometer metal oxide, toxicity

中图分类号: