Evaluation of the solvation parameter model as a quantitative structure-retention relationship model for gas and liquid chromatography

•Retention prediction error bench marked for the solvation parameter model in GC and RPLC•Average absolute error for retention factor predictions in GC 0.1 to 0.4•Average absolute error for retention factor predictions in RPLC 0.3 to 0.5•Descriptor quality for WSU compound descriptor database verified•Retention mechanism heterogeneity is the main source of prediction errors The Wayne State University (WSU) experimental descriptor database is utilized to bench mark the current capability of the solvation parameter model for use as a quantitative structure-retention relationship tool for estimating retention in gas and reversed-phase liquid chromatography. The prediction error for the retention factors of varied compounds on six open-tubular columns for gas chromatography (Rtx-5 SIL MS, DB-35 ms, RtxCLPesticides, HP-88, HP-INNOWAX and SLB-IL76) and three packed columns for reversed-phase liquid chromatography (SunFire C18, XBridge Shield RP18, and XBridge Phenyl) is used to establish expectations related to current practices. Each column data set was divided into a training set for calibration and a test set for validation employing a roughly 1 to 2 split, such that each test set contained about 40 to 80 varied compounds. The average absolute error for the prediction of retention factors by gas chromatography varied from about 0.1 to 0.4 on the retention factor scale with the larger error typical of stationary phases ranked as the most polar (or cohesive). For reversed-phase liquid chromatography the average error for the prediction of retention factors was 0.3 to 0.5 and generally larger than for gas chromatography. Statistical filters where utilized to identify a group of polycyclic aromatic compounds without hydrogen-bonding functional groups with a larger prediction error on the SunFire C18 column than for other compounds of smaller size, flexible structure or containing hydrogen-bonding functional groups. The heterogeneity of the retention mechanism is speculated to be the main contribution to the prediction error for both gas and liquid chromatography using the solvation parameter model.