INTRODUCTION Flavonoids are characterized by a common 2-phenyl-benzopyran-4-one basic structure and constitute one of the largest groups of naturally occurring compounds (Figure 1)¹. These compounds have been reported to display a variety of biochemical properties including antioxidant², antimicrobial and pharmaceutical activities 3,4 . They are also used as anti-inflammatory, antiviral, antiallergic, antibiotic, and anticarcinogenic compounds 3,5 . One of the most interesting biological properties of flavonoids is their ability to inhibit human immunodeficiency virus (HIV) transcriptase and HIV replication 6 . The diversity in molecular architecture for Oriental Journal of Chemistry Vol. 25(1), 49-56 (2009) Application of neural network to quantitative structure anti-HIV activity relationships of flavonoid compounds Y. BELMILOUD 1 , A. KADARI 1 , L. BENAHMED 2 , D. CHERQUAOUI 3 , D. VILLEMIN 4 and M. BRAHIMI 1 ¹Laboratoire de Physico-Chimie Theorique et de Chimie Informatique, Faculte de Chimie, U.S.T.H.B., BP 32 Al-alia ; Bab-Ezouar ; Alger (Algeria). ²Universite de Tlemecen Faculté de chimie, Tlemcen (Algerie). ³Departement de Chimie, Faculte des Sciences Semlalia BP 2390 Université Cadi Ayyad, Marrakech (Morocco). 4 Ecole Nationale Supérieure d’Ingénieurs (ENSICAEN) LCMT, UMR CNRS 6507, 6 boulevard Maréchal Juin, 14050 Caen Cedex (France). (Received: December 10, 2008; Accepted: January 15, 2009) ABSTRACT Artificial neural network (NN) was constructed and trained for the prediction of the anti- human immunodeficiency virus (anti-HIV) activity for 26 flavonoîd compounds based on quantitative structure- activity relationship method (QSAR). For different models, The network, inputs were selected by the stepwise multiple linear regressions technique (MLR) by using Codessa program.NN based obtained results lead to statistical results in good agreement with the literature data. They put in evidence the importance of the molecular hydrophobicity, electronegativity and atomic charges on some key atoms in modelling flavonoid compounds' behaviour by means of QSAR approach. Nonlinear NN models are shown to give better results with good predictive anti-HIV activity than linear ones. Key words: Flavonoid, multiple linear regressions technique MLR, quantitative structure-activity relationship method QSAR, neural network NN, anti-HIV, DFT. flavonoid compounds has made possible the development of different quantitative structure- activity relationships (QSAR), allowing the identification of molecular parameters responsible for their biological and physicochemical properties 7 . To understand the chemical mechanisms associated with the biochemical effect of flavonoid, various QSAR studies have been employed to research statistical relationships between molecular structure-derived parameters and the anti-HIV properties of flavonoids 8-11 . Anti-HIV activity for 26 flavonoid compounds¹², has been yet investigated in statistical analysis, by Multiple Linear Regression (MLR) method. The best regression equation obtained by