论文部分内容阅读
有机化合物的水溶解度是重要的物理化学性质。本文用18个拓扑符去描述1 293个化合物的分子结构。先建立1 293个有机化合物的分类模型,按照logS的大小,将数据分成三类,通过训练集建立模型,并用测试集检验,分类准确率达92.2%。在此基础上,以上述18个描述符作为输入,logS作为输出,研究水溶解度的定量,建立支持向量机预测模型。比较测试集的结果,以前建立的人工神经网络模型相关系数r~2=0.94和标准偏差sd=0.52,而本文建立的支持向量机模型r~2=0.95和sd= 0.50,显然优于以前建立的模型。
The water solubility of organic compounds is an important physicochemical property. In this paper, 18 topological symbols are used to describe the molecular structure of 1 293 compounds. Firstly, the classification model of 1 293 organic compounds was established. According to the size of logS, the data were divided into three categories. The model was set up by training set and tested by test set. The classification accuracy rate was 92.2%. Based on this, we use the above 18 descriptors as input and logS as output to study the quantitative of water solubility and establish the support vector machine prediction model. Compared with the result of the test set, the correlation coefficient r ~ 2 = 0.94 and the standard deviation sd = 0.52 of the previously established artificial neural network model, while the support vector machine model r ~ 2 = 0.95 and sd = 0.50 established in this paper is obviously better than the previous established Model.