Quantitative structure-toxicity relationship: An "in silico study" using electrophilicity and hydrophobicity as descriptors

To investigate the importance and suitability of quantitative structure-toxicity relationship approach in the field of aquatic toxicology, we have performed an extensive study introducing multiple linear regression (MLR) and multilayer perceptron neural network (MLP-NN) techniques. In this study, toxicity (pIGC(50)) prediction of 169 aliphatic compounds toward Tetrahymena pyriformis (a freshwater protozoa) has been made by using all possible combinations of electrophilicity index (omega), square of electrophilicity index (omega(2)), cube of electrophilicity index (omega(3)), hydrophobicity (logP), and its square term {(logP)(2)} as predictors in the developed models. The MLR and MLP employed to construct the linear prediction models for the complete sets lead to a good correlation coefficient (R-2) ranging from 0.703 to 0.779 in case of electronic factors (omega, omega(2), omega(3)) and 0.790 to 0.983 in case of lipophilic factors {logP, (logP)(2)}, respectively, except for amino alcohols. Furthermore, to cross-check the variable selection, a three-set cross-validation approach has been carried out. To demonstrate our overall result, the sum of ranking differences with ties has been evaluated considering the whole data set.