Comparison Study of Peptide Retention Time Prediction Model Based on Five Kinds of Amino Acid Descriptors in HPLC by Support Vector Machine

Abstract

Based on amino acid descriptors(z-scales, c-scales, ISA-ECI,MS-WHIM and PRIN) and additive method, evaluation of predict performance of five amino acid descriptors in peptide QSRR(Quantitative structure-retention relationships) with 101 promiscuous peptides in High-Performance Liquid Chromato- graphy by support vector regression(SVR) is made in the article, and RBF(radical basis function) is selected as kernel function. Using leave-one-out cross-validation (LOO-CV), we suppose that predicting accuracy of ISA-ECI is better than the other descriptors in SVR with RBF. The prediction correlation coefficient of the SVR model (ε = 0.001,σ= 5 and C= 100) is 0.8445 by leave-one-out cross validation. The standard error of prediction (SEP) error of the dataset is 1.03 by fitting calculation, and the prediction correlation coefficient is 0.9642.The prediction results are in agreement with the experimental values. This paper provided a simple and effective method for predicting the retention behavior of peptide and some insight into what structural features are related to the retention time of peptides. Moreover, it also offered an idea about nonlinear relation between retention time of peptides and their structural descriptors (ISA-ECI).Therefore, SVR is assumed to be a feasible method in peptide QSAR (Quantitative structure-activity relationships) model.