The molecular electronegativity distance vector（MEDV） was applied to characterize the molecular structures of 30 organophosphorous compounds. Optimum MEDV descriptors were selected by using the variable selection an...The molecular electronegativity distance vector（MEDV） was applied to characterize the molecular structures of 30 organophosphorous compounds. Optimum MEDV descriptors were selected by using the variable selection and modeling method based on the prediction（VSMP） technique. The quantitative structure-toxicity relationship（QSTR） model was built for acute toxicity（96h pLC50） of organophosphorous compounds to steelhead. The developed QSTR model with strictly internal and external validations presents relatively high correlation coefficient（R2） of 0.9518, leave-one-out（LOO） cross-validated correlation coefficient（Q2LOO） of 0.9355, and leave-many-out（LMO） cross-validated correlation coefficient（Q2LMO） of 0.9290. The robustness of the model was confirmed by the y-randomization test（R2yrand = 0.0772 and Q2 yrand = –0.5313） and bootstrapping（R2bstr = 0.9502 and Q2 bstr = 0.9177） method. The result of external validation, Q2F1 = 0.9336, Q2F2 = 0.9336, Q2F3 = 0.9447, r2 m = 0.8120, and CCC = 0.9602, shows that the QSTR model has a high predictive ability.展开更多
基金the financial support from the National Natural Science Foundation of China(21207024 and 21407032)the Provincial Natural Science Foundation of Guangxi(2014GXNSFAA118060,2014GXNSFBA118233 and 2013GXNSFBA019228)
文摘The molecular electronegativity distance vector（MEDV） was applied to characterize the molecular structures of 30 organophosphorous compounds. Optimum MEDV descriptors were selected by using the variable selection and modeling method based on the prediction（VSMP） technique. The quantitative structure-toxicity relationship（QSTR） model was built for acute toxicity（96h pLC50） of organophosphorous compounds to steelhead. The developed QSTR model with strictly internal and external validations presents relatively high correlation coefficient（R2） of 0.9518, leave-one-out（LOO） cross-validated correlation coefficient（Q2LOO） of 0.9355, and leave-many-out（LMO） cross-validated correlation coefficient（Q2LMO） of 0.9290. The robustness of the model was confirmed by the y-randomization test（R2yrand = 0.0772 and Q2 yrand = –0.5313） and bootstrapping（R2bstr = 0.9502 and Q2 bstr = 0.9177） method. The result of external validation, Q2F1 = 0.9336, Q2F2 = 0.9336, Q2F3 = 0.9447, r2 m = 0.8120, and CCC = 0.9602, shows that the QSTR model has a high predictive ability.
