贝叶斯规整化神经网络预测苯取代物的毒性

利用贝叶斯规整化神经网络(BRNN)研究了59种苯取代的环境毒性和分子结构的关系.建立了基于脂水分配系数和4种量子指数的QSAR环境毒性预测模型.结果表明:该模型的预测性能超过偏最小二乘法(PLS)和逐步线性回归方法,残差分析显示其稳定可靠,有望成为一种良好的苯取代物毒性的环境评价和预测的方法.

Toxicity Prediction for Substituted Benzenes with Bayesian Regularized Neural Network

The structure-toxicity relationship of 59 substituted benzenes was studied by the method of bayesian regularization neural network(BRNN),and a QSAR model was built for environmental toxicity evaluation and prediction of substituted benzenes with lipid-water partition coefficient and four quantum indices.The results of cross validation indicated that the prediction ability of this model are much better than the stepwise MLR and PLS model,and residual analysis shows that the model is robust and stable.This model is proved to be an effective method for the evalution and prediction of the environmental toxicity of substituted benzenes.