Tarko L.,Nenitzescu of Romanian Academy |
Putz M.V.,West University of Timișoara
Journal of Theoretical and Computational Chemistry | Year: 2012
This paper presents result of QSTR (quantitative structuretoxicity relationship) study obtained using the PRECLAV software. The dependent property is toxicity against rat (Rattus norvegicus), measured by TD 50 values. The calibration/training/learning set includes 49 molecules having a very high chemical diversity. There are five outliers in calibration set. In the presence of outliers the predictive power of QSTR is very low (r 2 = 0.5425, F = 10.4, r 2 CV = 0.4387). After elimination of outliers the predictive power of QSTR is much higher (r 2 = 0.9078, F = 44.3, r 2 CV = 0.8776). All eight predictors are nonlinear functions (parabolic and products) of descriptors. The LogP is not predictor. Presence of C = CH 2 and N-NO molecular fragments increases toxicity. Presence of C 6H 4 molecular fragment decreases toxicity. © 2012 World Scientific Publishing Company.