A Novel QSAR Combination Forecast Model for prediction of Bioactivity of Aromatic Carboxylic Acid Derivatives Insect Repellent Based on Support Vector Regression and K-Nearest-Neighbor

摘要

In this work a novel nonlinear combination forecast model was proposed based on support vector regression (SVR) and AT-nearest neighbor (KXN) to improve the precision of quantitative structure-activity relationship (QSAR) modeling for aromatic carboxylic acid derivatives insect repellent: Firstly,search optimal kernel function and select molecular descriptors nonlinearly by the rule of minimum MSE value using SVR.Secondly,illuminate the effect of all descriptors on biological activity by "multi-round enforcement resistance-selection".Thirdly,construct sub-models with predicted values of different KNN.Then,get optimal kernel and corresponding retained sub-models by screening submodels.Finally,make prediction based on retained submodels with leave-one-out (LOO) method.Multiple linear regression (MLR) and stepwise linear regression (SLR) were also utilized to construct the linear QSAR models for insect repellent The results using SVR-KNN were compared with MLR,SLR and previous published model and indicate the superiority of the present combination forecast model.