Modeling GC-ECD retention times of pentafluorobenzyl derivatives of phenol by using artificial neural networks

The depicted retention times (RTs) of an electron capture detection (ECD) system is predicted for a set of 37 pentafluorobenzyl (PFB) derivatives of phenol in a semi-polar column, DB-1701 (14% cyanopropylphenyl and 86% dimethyl-polysiloxane). Among a large number of descriptors, four parameters categorized as electronic, topological, geometric, and hybrid (geometric and topological) descriptors are chosen using stepwise multiple regression technique. Each molecular descriptor in this model was disputed to unfold the relationship between molecular structures and their RTs. The descriptors occurring in the multiple linear regression (MLR) model were considered as inputs for developing the back propagation artificial neural networks (BPANN). The artificial neural network (ANN) model shows superiority over the MLR by decerning 91.9% for different classes of molecules in confusion matrix. This refers to the fact that the retention behaviors of molecules display nonlinear characteristics. The accuracy of 4-4-1 BPANN model was illustrated using leave-one-out (LOO), leave-multiple-out (LMO) cross-validations, and Y-randomization. Moreover, the mean effect of descriptors betrays that descriptor Ss is the most indispensable factor affecting the retention behavior of molecules.