Prediction of electrophoretic mobilities of peptides in capillary zone electrophoresis by quantitative structure-mobility relationships using the Offord model and artificial neural networks

The aim of this work was to explore the usefulness of empirical models and multivariate analysis techniques in predicting electrophoretic mobilities of small peptides in capillary zone electrophoresis (CZE). The data set consists of electrophoretic mobilities, measured at pH 2.5, for 125 peptides ranging in size between 2 and 14 amino acids. Among the existing empirical models, the Offord model (i.e., mu identical with Q/M(2/3)) gave the best correlation for the data set. A quantitative structure-mobility relationship (QSMR) was developed using the Offord's charge-over-mass term (Q/M(2/3)) as one descriptor combined with the corrected steric substituent constant (E(s, c)) and molar refractivity (MR) descriptors to account for the steric effects and bulkiness of the amino acid side chains. The multilinear regression (MLR) of the data set showed an improvement in the predictive ability of the model over the simple Offord's relationship. A 3-4-1 back propagation artificial neural networks (BP-ANN) model resulted in a significant improvement in the predictive ability of the QSMR over the MLR treatment, especially for peptides of higher charges that contain basic amino acids arginine, histidine, and lysine. The improved correlations by the BP-ANN analysis suggest the existence of nonlinear characteristic in the mobility-charge relationships.