Abstract

QSAR study on a data set of 5-lipoxygenase inhibitors (1-phenyl [2H]-tetrahydro-triazine-3-one analogues) was carried out by using Support Vector Regression (SVR) and physicochemical parameters. Wrapper methods were used to select descriptors, while Leave-One-Out Cross Validation (LOOCV) method and independent set test were used to judge the predictive power of different models. We found out that the generalization ability of SVR model outperformed multiple linear regression (MLR) and Partial Least Squares (PLS) models in this work. An online web server for activity prediction is available at http://chemdata.shu.edu.cn/qsar5lip.

Keywords: Support vector regressions, 1-phenyl [2H]-tetrahydro-triazine-3-one analogues, Leave-one-out cross-validation, Feature selection, Wrapper, Multiple linear regression (MLR), Partial least squares (PLS) analysis