Prediction of protein retention times in anion-exchange chromatography systems using support vector regression

Quantitative Structure-Retention Relationship (QSRR) models are developed for predicting protein retention times in anion exchange chromatography. Constitutional, topological and electron-density-based descriptors are computed directly from the protein crystal structures. These QSRR models are constructed based on the Support Vector Regression (SVR) algorithms. To accomplish this, a two-step computational strategy was adopted. In the first step, a linear SVR was utilized as a variable selection method and the relative importance of selected descriptors is analyzed using the star plot visualization approach. Subsequently, the selected features are used to produce nonlinear SVM bagged models. After validation, these predictive models may be used as part of an automated virtual high-throughput screening (VHTS) process.