Quantitative Structure Activity/Pharmacokinetics Relationship Studies Of HIV-1 Protease Inhibitors

Highly active antiretroviral therapy?HAART?is the common adopted clinical treatment for ADIS patient at present,as HIV-1 protease inhibitor?PIs?is one important component of HAART,thus,it is of significant importance to study HIV-1PIs.Quantitative structure activity/pharmacokinetics relationship?QSAR/QSPR?studies areeffectivetoolsforshortening the drugdevelopmenttime.Considering the structure diversity of PIs,support vector machine?SVM?,partial-least squares regression?PLSR?and back-propagation neural network?BPNN?were applied to study the QSAR/QSPR in this paper.In this study,five steps were carried out to screen out descriptors that generated from the E-dragon at VCCLAB.In the end,seven descriptors that relevant with the dissociation rate(k_off)and five descriptors with the inhibitory constant?K_i?remained in the end.This paper find that three descriptors that relevant with the K_i were all belong to the 3D-MoRSE class,which was calculated by the angle scattering function.This finding indicates that K_i may have a closer link with the 3D-MoRSE descriptor,and k_offff is likely to have more affective factors.For the QSAR models,SVM,PLSR and BPNN all generated reliable prediction models with the non-cross validation cofficient?r²?of 0.688,0.768 and 0.787,respectively,and non-cross validation coefficient of the test set??? of 0.748,0.696and 0.640,respectively.The kernel function of the best SVM model is linear function,while that of the best BPNN is trains.For the QSPR models,the optimum models of SVM,PLSR and BPNN got the r² of 0.952,0.869 and 0.960,respectively,and the ??? of 0.852,0.628 and 0.814,respectively.The kernel function of the best SVM model is radial basis functions,while that of the best BPNN is traindx.The final statics indicate that these three methods all get effective models,and among these three modeling methods,SVM showed superior ability to PLSR and BPNN both in QSAR/QSPR modeling of PIs,thus,we suspected that SVM was more suitable for predicting activities of PIs and complex kernel function is more suitable for predicting complex parameter.In addition,we also found that 3D-MoRSE descriptors may have a tight relationship with the K_i values of PIs,and the GETAWAY descriptors have significant influence for both k_offff and K_i from PLSR equations.In one word,I believe that this work can provide theoretical basis for the future HIV-1 PIs designing.