Collect. Czech. Chem. Commun. 2009, 74, 217-241
Published online 2009-02-11 11:38:15

A comparative QSAR study of SVM and PPR in the correlation of lithium cation basicities

Alan R. Katritzkya,*, Yueying Rena, Svetoslav H. Slavova and Mati Karelsonb,c

a Center for Heterocyclic Compounds, Department of Chemistry, University of Florida, Gainesville, Florida 32611, USA
b Department of Chemistry, University of Tartu, 2 Jakobi Street, Tartu 51014, Estonia
c Department of Chemistry, Tallinn University of Technology, Ehitajate tee 5, Tallinn 19086, Estonia


Correlation of gas-phase lithium cation basicities (LCB) of 259 diverse compounds extends the published datasets utilizing multilinear, support vector machine (SVM) and projection pursuit regression (PPR) modeling. The best multiple linear regression (BMLR) method implemented in CODESSA was used to: (i) build multiparameter linear QSPR models and (ii) select set of descriptors for further treatment by the SVM and PPR. The external predictivity and the performance of each of the above methods was estimated and compared to those of the other techniques. The PPR method produced results superior to SVM, which in turn outperformed MLR. The physico-chemical interpretation of each of the descriptors provides new insight into the mechanism of LCB interactions.

Keywords: Quantitative structure-property relationships; Lithium cation basicity; CODESSA; Multiple linear regression; Support vector machines; Projection pursuit regression.

References: 40 live references.