2001/05/12 by Н.С. Зефиров, Vladimir A. Palyulin · 1 voice · 1 citation
Chemistry · Computer Science · Mathematics · #Computational Drug Discovery Methods #Free Radicals and Antioxidants #Graph theory and applications
paper · doi:10.1021/ci0001637
openalex publication_date 2001/05/12 · openalex created_date 2025/10/10 · openalex updated_date 2026/06/16
Our investigation was motivated by a recent paper concerning "high-quality structure-property-activity regressions" (J. Chem. Inf. Comput. Sci. 2000, 40, 899-905) for boiling points of small sulfides using variable connectivity indices. We performed QSAR study of the same data set using a logically preselected solvation index and obtained a very good one-parameter regression. The structures of the whole possible set of small sulfides (C2-C6) were generated and the statistics were proven by real prediction using an external test set of sulfides. The variants of extended prediction with extrapolated data and QSAR using an expanded training set were also performed, and all these data also revealed the preference of the solvation index. A general problem of descriptive vs predictive QSAR is discussed.