基于基团贡献法的有机化合物好氧生物降解预测模型研究

Group-Contribution-Method-Based Model for Prediction of Aerobic Biodegradation of Organic Compounds

从MITI-Ⅰ试验中筛选出587种不同类型有机化合物的可用数据,通过对这些物质的结构进行拆分,随机选择其中50种化合物作为验证集,另外537种作为训练集,利用多元线性回归(MLR)和支持向量机(SVM)2种计算方法分别建立模型。结果表明,芳香酸、醛、芳香碘和叔胺等功能基团对有机化合物的好氧生物降解性影响较大;MLR模型总体预测正确率为81.43%,验证集正确率为82%,SVM模型总体预测正确率为87.90%,验证集正确率为86%。所建立的2种定量结构与生物降解性关系(QSBR)模型有效,可用于化学品的好氧生物降解性评价。

From MITI- 1 test, a total of 587 different kinds of organic compounds were screened out. Through structure splitting, 50 kinds of these were picked out randomly as a validation set, and the other 537 as a training set. Models were built up using the multiple linear regression method （MLR） and support vector machine （SVM） , separately, for predic- ting aerobic biodegradation of organic chemicals. Results show that functional groups, such as aromatic acid, aldehyde, aromatic iodine and tetriary amine, had much effect on aerobic biodegradability of these organic compounds. The predic- tion using the multiple-linear-regression-based model reached 81.43% in overall accuracy, and 82% with the validation set ; and the use of the other model based on support vector machine reached 87.90% in overall accuracy, and 86% withthe validation set. The two kinds of QSBR models can be used to evaluate aerobic biodegradability of organic chemicals.