论文部分内容阅读
本文选取了部分有机物致敏性和部分有机物极性参数两组数据,均采用ADMEWORKS ModelBuilder软件计算并选择出合适的结构描述符,进而采用K最近邻和K均值聚类法对两组数据进行分类,然后对分类后的数据分别运用多元线性回归(Multiple Linear Regression,MLR)、偏最小二乘(Partial Least Squares,PLS)和人工神经网络(Artificial Neural Networks,ANN)方法进行QSPR建模研究。结果表明,无论采用何种分类方法都可以在一定程度上改善模型预测的结果。对于两组样本,有机物分子结构差异较小的样本集模型预测结果较优,非线性模型的预测结果整体优于或相当于线性模型的预测结果。
In this paper, we selected two sets of data of the sensitivities of some organic compounds and the polar parameters of some organic compounds. We used ADMEWORKS ModelBuilder software to calculate and select the appropriate structure descriptors, and then classified the two sets of data using K nearest neighbor and K means clustering , And then classified the data using multiple linear regression (MLR), Partial Least Squares (PLS) and Artificial Neural Networks (ANN) methods QSPR modeling studies. The results show that no matter what classification method is used, the result of model prediction can be improved to some extent. For the two groups of samples, the sample set model with smaller difference of molecular structure of organic substance has better prediction result, and the prediction result of the nonlinear model is better than or the same as the linear model.