Du Hongying, Wang Jie, Hu Zhide, Yao Xiaojun, Zhang Xiaoyun
Department of Chemistry, Lanzhou University, Lanzhou, China.
J Agric Food Chem. 2008 Nov 26;56(22):10785-92. doi: 10.1021/jf8022194.
Three machine learning methods, genetic algorithm-multilinear regression (GA-MLR), least-squares support vector machine (LS-SVM), and project pursuit regression (PPR), were used to investigate the relationship between thiazoline derivatives and their fungicidal activities against the rice blast disease. The GA-MLR method was used to select the most appropriate molecular descriptors from a large set of descriptors, which were only calculated from molecular structures, and develop a linear quantitative structure-activity relationship (QSAR) model at the same time. On the basis of the selected descriptors, the other two more accurate models (LS-SVM and PPR) were built. Both the linear and nonlinear modes gave good prediction results, but the nonlinear models afforded better prediction ability, which meant that the LS-SVM and PPR methods could simulate the relationship between the structural descriptors and fungicidal activities more accurately. The results show that the nonlinear methods (LS-SVM and PPR) could be used as good modeling tools for the study of rice blast. Moreover, this study provides a new and simple but efficient approach, which should facilitate the design and development of new compounds to resist the rice blast disease.
采用三种机器学习方法，即遗传算法-多元线性回归（GA-MLR）、最小二乘支持向量机（LS-SVM）和投影寻踪回归（PPR），研究噻唑啉衍生物与其对稻瘟病的杀菌活性之间的关系。GA-MLR方法用于从仅根据分子结构计算得到的大量描述符中选择最合适的分子描述符，并同时建立线性定量构效关系（QSAR）模型。基于所选描述符，构建了另外两个更精确的模型（LS-SVM和PPR）。线性和非线性模式均给出了良好的预测结果，但非线性模型具有更好的预测能力，这意味着LS-SVM和PPR方法能够更准确地模拟结构描述符与杀菌活性之间的关系。结果表明，非线性方法（LS-SVM和PPR）可作为研究稻瘟病的良好建模工具。此外，本研究提供了一种新的、简单但高效的方法，这应有助于抗稻瘟病新化合物的设计与开发。
