Title: Classification Structure-Activity Relationship Study of Reverse Transcriptase Inhibitors
Volume: 8
Issue: 7
Author(s): Maria Seyagh, EL Mostapha Mazouz, Andreea Schmitzer, Didier Villemin, Abdellah Jarid and Driss Cherqaoui
Affiliation:
Keywords:
Artificial neural networks, Decision trees, Linear discriminant analysis, Support vector machines, Structure-activity relationships, Acquired immuno-deficiency syndrome (AIDS), T-lymphocytes, proviral DNA, RT inhibitors, non-nucleoside inhibitors (NNRTI)., tetrahydroimidazo[4,5,1-jk][1,4]benzodiazepinone (TIBO), retroviruses, human immunodeficiency virus type 2 (HIV-2), multiple linear regression (MLR), anti-HIV HEPT derivatives
Abstract: A classification structure – activity relationship study has been carried out using topological indices, physicochemical and steric parameters on a series of 1-[2-hydroxyethoxy-methyl]-6-(phenylthio) thymine for their HIV reverse transcriptase inhibitory activity. The predictive classification performance of support vector machines method is investigated and compared with those of other classifiers such as artificial neural networks, linear discriminant analysis, k-nearest neighbours and decision trees. This paper discusses several validation strategies including randomization test, internal and external validations. The quality of the models was evaluated by the number of right classified compounds. The results obtained show that all methods used except k-nearest neighbours were good classifiers. The percentage of right classified compounds ranges from 87.7% to 95.4% and from 64.3% to 92.9% for the training and test sets, respectively. The relevant factors controlling the anti-HIV activity have been identified. The descriptors related to both steric characters (6Xvch, 4XNP and 1/S) and hydrophobic parameter (logP) seem to be very relevant in the establishment of structure-anti- HIV activity relationship.
