化学品厌氧生物降解QSBR预测模型研究  

QSBR Models for Predicting Anaerobic Biodegradation of Chemicals

作　　者：马益[1,2] 刘济宁[1,2] 陈英文[1] 石利利[2] 沈树宝[1] 

机构地区：[1]南京工业大学生物与制药工程学院,江苏南京210009 [2]环境保护部南京环境科学研究所,江苏南京210042

出　　处：《生态与农村环境学报》2012年第3期310-314,共5页Journal of Ecology and Rural Environment

基　　金：国家高技术研究发展计划(2009AA05Z313);国家自然科学基金(50872052;1172107);国家自然科学基金青年科学基金(21106072);公益性行业(环保)科研专项(200909086);江苏省环保厅项目(2009002);江苏省科技厅项目(BE2010196);江苏高校优势学科建设工程项目

摘　　要：收集了155种有机化学品厌氧生物降解数据,以随机抽取的109种物质作为训练集,另外46种物质作为验证集,通过结构式拆分得到各基团,分别采用多元线性回归和BP人工神经网络2种算法对有机化合物结构与生物降解性定量关系(QSBR)进行研究。结果表明,多元线性回归模型验证集正确率为78.26%,总正确率为84.52%;BP人工神经网络模型验证集正确率为82.61%,总正确率为90.32%。可见,BP人工神经网络算法相对优于多元线性回归算法。Anaerobic biodegradation data of 155 kinds of organic chemicals were collected. Among them the data of 109 chemicals were picked up randomly for use in the training set, and of the remaining 46 in the validation set. By splitting constitutional formula, radicals or groups of the chemicals were obtained and analyzed with muhiple linear regression and BP artificial neural network, separately, to explm＇e for quantitative relationships between structure and biodegradability of organic chemicals （QSBR）. Results show that the use of MLR in analyzing the validation set was 78.26% in accuracy and 84. 52% in total accuracy and for the use of BP-ANN, 82.61% in accuracy and 90. 32% in total accuracy. Obviously the latter is superior to the former.

关 键 词：厌氧生物降解 人工神经网络 多元线性回归 QSBR 

分 类 号：X131[环境科学与工程—环境科学] X327