Selection of calibration compounds for selectivity evaluation of siloxane-bonded silica columns for reversed-phase...

For the faster evaluation of selectivity in reversed-phase liquid chromatography of siloxane-bonded silica columns using the solvation parameter model a minimal set of calibration compounds is described suitable for mobile phase composition from 20-70% (v/v) methanol-, acetonitrile-, or tetrahydrofuran-water. The Kennard-Stone uniform mapping algorithm is used to select the calibration compounds from a larger database of compounds with known retention properties used earlier for column selectivity evaluation. Thirty-five compounds are shown to be necessary to minimize the standard deviation of the system constants and to minimize the difference between the system constants determined by conventional calibration and the values obtained for the reduced calibration compounds. The models for SunFire C₁₈ with methanol-, acetonitrile- and tetrahydrofuran-water mobile phase compositions and XBridge Shield RP18, XBridge C₈, XBridge Phenyl and Discovery HS F5 with methanol- and acetonitrile-water mobile phase compositions had an average coefficient of determination of 0.996 (standard deviation = 0.003, n = 11) and average standard error of the estimate 0.025 (standard deviation = 0.005, n = 11) for the reduced calibration compounds. Some octadecylsiloxane-bonded silica stationary phases with a high bonding density and methanol-water mobile phase compositions containing ≤ 30% (v/v) methanol exhibit extreme retention factors (log k > 2.5) for the low-polarity, two-ring aromatic compounds in the thirty-five compound calibration set. Alternative calibration compounds with more favorable retention properties are suggested as replacements in these cases. The predictive capability of the calibration models is validated using external test sets characterized by the average error, average absolute error and root mean square error of prediction. For the thirty-five calibration compounds sets the average absolute error 0.026 (standard deviation = 0.009, n = 11) and root mean square error of prediction 0.032 (standard deviation = 0.010, n = 11) confirm the suitability of the calibration models for column selectivity evaluation. System maps for XBridge Shield RP18 for 20-70% (v/v) methanol-water and Synergi Hydro-RP and 50% (v/v) methanol-water at temperatures from 25-65 °C together with a correlation diagram for XBridge Shield RP18 and SunFire C₁₈ are presented as representative applications of the reduced calibration compounds for column selectivity evaluation.