144 O. Tarasova, A. Lagunin, D. Filimonov, V. Poroikov V.N. Orekhovich Institute of Biomedical Chemistry of RAMS A. Zakharov National Cancer Institute, National Institute of Health M. Stasevych, V. Zvarych, R. Musyanovych, V. Novikov Institute of Chemistry and Chemical Technology, Lviv Politechnic National University (Q)SAR analysis of anthraquinone and naphtoquinone derivatives Despite some published data on phytotoxic and antifungal action of several naphtoquinone derivatives, biological properties of these two chemical classes are not studied in detail. Thirty three new anthraquinone derivatives and 61 naphtoquinone derivatives were synthesized and tested against several types of biological activities including antibacterial, antifungal, antitumor, antiischemic, anticonvulsant and neuroprotector action. (Q)SAR analysis of new anthraquinone and naphtoquinone derivatives was performed based on their data on antibacterial and antifungal activities . (Q)SAR approach implemented in GUSAR program has been applied to the creation of (Q)SAR models and testing them in leave-many-out procedure. The minimal accuracy obtained in leave 30% out procedure was 85,8% for antibacterial activity of naphtoquinone derivatives and 100% for antibacterial and antifungal activities of anthraquinone derivatives. Based on the obtained (Q)SAR models several compounds were selected for the subsequent synthesis and testing. A novel (Q)SAR approach based on the self-consistent regression implemented in GUSAR program was applied to the creation of (Q)SAR models [1]. The latest version of GUSAR (2013.1) allows creating category-based models. Anthraquinone and naphtoquinone derivatives tested against antibacterial and antifungal activities were used for the models creation. Other studied activities were excluded from the consideration due to the lack of compounds for the creation of models. Models were created for anthraquinone derivatives and naphtoquinone derivatives separately. Training sets consisted of 12 anthraquinone derivatives tested against antibacterial and antifungal activities (7 active and 5 inactive molecules) and 32 naphtoquinone derivatives tested against antibacterial activity (19 active and 13 inactive molecules) and 32 molecules – against antifungal activity (20 active molecules and 12 inactive molecules). Taking into account small amount of compounds in the training set, we decided to generate 100 different models and then select the best models based on the accuracy obtained in leave- many-out procedure (LMO). The best models were used to create the consensus model. Since the amount of antrhaquinone derivatives in the training set is very small, leave 20% out procedure has been performed (leaving more compounds out of training set would be impossible due to the restrictions of GUSAR – number of compounds in the training set should exceed 10). Leave 30% out procedure had been performed for the training set of naphtoquinone derivatives. The characteristics of consensus (Q)SAR models are given in the Table 1. The best results in LMO could be obtained without taking into account K nearest neighbors in the set of anthraquinone derivatives and taking into account K nearest neighbors in the set of naphtoquinone derivatives. Table 1. The characteristics of (Q)SAR consensus models obtained in leave 20% out* and leave 30% out** procedures. Sensitivity Specificity Balanced accuracy Mean accuracy Anthraquinones L20Out* L20Out L20Out L20Out Antibacterial 0.947 0.769 0.858 0.854 Antifungal 1.0 0.917 0.958 0.957 Naphtoquinones L30Out** L30Out L30Out L30Out Antibacterial 1.0 1.0 1.0 1.0 Antifungal 1.0 1.0 1.0 1.0 Lviv Polytechnic National University Institutional Repository http://ena.lp.edu.ua