Prediction of retention times of peptides in RPLC by using radial basis function neural networks and projection pursuit regression

Abstract

Quantitative structure–retention relationship (QSRR) models correlating the retention times of peptides in reversed-phase liquid chromatography (RPLC) and their structures were developed based on linear and non-linear modeling methods. The best multi-linear regression (BMLR) method implemented in the CODESSA was used to select the most appropriate molecular descriptors from a large set of descriptors and develop a linear QSRR model. Using the selected descriptors, another two non-linear regression methods (radial basis function neural networks (RBFNN) and projection pursuit regression (PPR)) were used in the non-linear QSRR models development. The predicted retention times from the two non-linear approaches RBFNN and PPR were in good agreement with the experimental data. The coefficients of determination (R 2) for the training set of these two methods (RBFNN and PPR) were 0.9787 and 0.9881; the root mean square of errors (RMSE) of these two methods were 0.5666 and 0.4207. They proved that RBFNN and PPR were very useful methods with good predictive ability for the prediction of peptides' RPLC retention times. The proposed methods will be of importance in the proteomic research, and could be expected to apply to other similar research fields.