Oliveira, K. M. G.; Takahata, Y. QSAR Modeling of Nucleosides Against Amastigotes of Leishmania donovani Using Logistic Regression and Classification Tree. QSAR Comb. Sci. 2008, 27, 1020–1027.

Abstract

We employed two classification methods; first, a logistic regression, second, classification tree, to classify nucleoside activities against Leishmania donovani using a training set of 21 compounds. The compounds are classified either active or inactive. The model was validated using a test set of 14 compounds. Two descriptors, Mor26v and Gap(HOMO, HOMO-1), were selected. The logistic regression resulted classification accuracy of 90.5% for the training set, 67% for the test set after Applicability Domain analysis was performed. The method of classification tree resulted classification accuracy of 95% for the training set, 86% for the test set. It was shown that the lowest energy conformation can be used to build a QSAR model through examination of the whole conformations that lie above the lowest energy conformation in the energy window of 13 kcal/mol. The selected descriptor Mor26v distinguishes differences in molecular chirality, while Gap(HOMO, HOMO-1) distinguishes differences in electronic structures.