Абстрактный

Application of multiple linear regression and artificial neural networks to predict LC50 in fish

Mehdi Alizadeh

A quantitative structure–activity relationship (QSAR) study has been carried out on 31 diverse organic pollutants by using molecular structural descriptors. Modeling of the logarithm values LC50 (lethal concentration required to kill 50%of a population) in fish (after 96 h) of these compounds as a function of the theoretically derived descriptors was established by multiple linear regression (MLR) and artificial neural networks (ANN). The Stepwise SPSS was used for the selection of the variables (descriptors) that resulted in the best-fitted models. For prediction logarithm values LC50 of compounds three descriptors were used to develop a quantitative relationship between the logarithm values LC50 and structural activity. Appropriate models with low standard errors and high correlation coefficients were obtained.After variables selection, compounds randomlywere divided into two training and test sets and MLR and ANN used for building the best models. The predictive quality of the QSAR models were tested for an external prediction set of 8 compounds randomly chosen from 31 compounds. The regression coefficients of prediction for training and test sets for ANN model were 0.9953 and 0.9938 srespectively. Result obtained showed that ANN model can simulate the relationship between structural descriptors and the Log LC50 of the molecules in data sets accurately and Theoretical predictions coincide verywell with experimental results.