A novel tree kernel support vector machine classifier for modeling the relationship between bioactivity and molecular descriptors

摘要

Support vector machine (SVM) has been gaining popularity in the field of chemistry. However, it also suffered from the problems of feature subset selection in most of applications. In the present study, we attempt to construct an informative novel tree kernel to address these problems. The constructed tree kernel can effectively discover the similarities of samples and handle nonlinear classification problems. Simultaneously, informative features can be evaluated by variable importance ranking in the process of building kernel by a large number of decision trees. Thus, under the framework of kernel methods, a novel tree kernel support vector machine (TKSVM) has been proposed to model the structure-activity relationship between bioactivities and molecular structures. Three datasets related to different categorical bioactivities of compounds are used to test the performance of TKSVM. The results show that the present method is a promising one compared to the SVM models with other commonly used kernels.