QSAR study of toxicity on fish based on molecular descriptors
Wang Heng,Li Yan,Ding Jun,Wang Yuan,Wang Yonghua,Chang Yaqing
DOI: https://doi.org/10.3969/j.issn.1001-4160.2009.04.020
2009-01-01
Abstract:The aim of this work is to predict the toxicity(LC_(50)) of the environmental compounds on fish based on QSAR methods,and to determine the key descriptors influencing the toxicity,and also to compare several mathematical methods in building models.The QSAR models were built based on a dataset of 114 diversified compounds.We randomly selected 75%(85 compounds) as training set and the rest 25%(29 compounds) as the test set.For each molecule,194 molecular indices were calculated.The models were developed using multiple linear regression(MLR),principal component analysis(PCA) and partial least squares(PLS),respectively.For the MLR model,the correlation coefficients R~2 between the experimental and predicted -logLC_(50) values of training set and test set are R_(tr)~2 =0.86 and R_(te)~2 =0.83,respectively,explaining that this model has the high reliability and robustness.For the PCA model,using 8 PCs,the correlation coefficients R~2 between the experimental and predicted -logLC_(50) values of training set and test set are R_(tr)~2 = 0.81 and R_(te)~2 =0.77;for the PLS model,using 5 latent components,the correlation coefficients R~2 between the experimental and predicted -logLC_(50) values of training set and test set are R_(tr)~2 =0.88 and R_(te)~2 =0.85.The key molecular structure parameters extracted by MLR which influence the toxicity on fish are electro-topological state indices(SssO,SsCl,SdCH_2,SsNH_2),molecular connectivity(Xv_0) and Kappa index(Ka_2).The present work should be valuable for evaluating the toxicity of environmental compounds on fish,as well as to deepen our insights to the toxic mechanism of compounds.