The influence of descriptor database selection on the solvation parameter model for separation processes.

• The WSU and Abraham compound descriptor databases are not interchangeable. • Strong correlation of descriptors but with reasonably large random differences. • The A descriptors are only weakly correlated. • System constants for mixed descriptor datasets linearly dependent on database source. The distribution of neutral compounds in biphasic separation systems can be described by the solvation parameter model using six solute properties, or descriptors. These descriptors characterize the size (McGowan's characteristic volume), V, excess molar refraction, E, dipolarity/polarizability, S, hydrogen-bond acidity and basicity, A and B, and the gas-liquid partition constant on n-hexadecane at 298.15 K, L. McGowan's characteristic volume and the excess molar refraction for liquids are available by calculation (E requires and experimental refractive index). The other descriptors and excess molar refraction for solids are experimental quantities and subject to greater variation or are estimated using computational or empirical models. Solute descriptors for several thousand compounds are available in the Abraham descriptor database and for several hundred compounds in the WSU descriptor database. These publicly accessible databases were developed independently using different approaches and for many compounds provide different descriptor values. In this report we evaluate the effect of mixing descriptors from the two databases on modeling chromatographic retention factors and liquid-liquid partition constants. It is shown that the two descriptor databases are not interchangeable. The WSU descriptor database consistently demonstrates improved model quality as determined by statistical parameters. Model system constants exhibit a general dependence on database selection with an approximately linear trend as a function of the fraction of compounds assigned descriptors from either database. There is no general model performance advantage to using mixed descriptor datasets and no real cause for concern for relatively large datasets containing < 15 % of compounds with descriptors assigned from the other database. For small datasets, descriptor quality is an important variable for adequate model performance. [ABSTRACT FROM AUTHOR]