Study on anti-fighting activity of mice by artificial neural network

Abstract

In order to study the quantitative structure-activity relationship (QSAR) of the anti-fighting activity E for 30 benzodiazepinooxazole derivatives against male mice, the molecular electronegativity distance vector M_D (D=1, 2, …, 91) of these compounds is calculated according to molecular topological environment. The three-variable (M₅₉, M₁₆, M₁₀) QSAR model of E for the compounds is constructed by stepwise regression method. The result demonstrates that the model is robust and has good prediction ability under <italic>R</italic>²_cv, <italic>V</italic>_IF tests. The M₅₉, M₁₆, M₁₀ are used as the input neurons of artificial neural network (ANN), and a 3∶5∶1 network architecture is employed. A satisfying BP-<italic>E</italic> model is constructed with the back-propagation algorithm, with the correlation coefficient R² and the standard error S being 0.984 and 0.054, respectively, showing that the relationship between E and the three structural parameters has a good nonlinear correlation. According to the three parameters of the model, it is clear that the dominant factors that impact the anti-fighting activity of benzodiazepineoxazole derivatives on male mice are the microscopic fragments: —CH₃, —CH₂—, , —NH— and —OH (O).