Prediction of LD50 for carboxylic acid derivatives using multiple linear regression and artificial neural networks models

In this research, Quantitative Structure–Activity Relationship (QSAR) study has been used for prediction of toxicity values of carboxylic acid derivatives. Firstly, the toxicity (LD50) values of data set of studied compounds were taken from the scientific web book and the their structures were drawn with the Gauss view 05 program and optimized at Hartree–Fock level of theory and 3-21G basis set by Gaussian 09 software. Then the dragon software was used for the calculation of molecular descriptors. The unsuitable descriptors were deleted with the aid of the genetic algorithm (GA) and backward techniques, and the best descriptors were used for multiple linear regression (MLR) and artificial neural network (ANN) models. The prediction accuracy of the final model was discussed using the statistical parameters. Leave-one-out cross-validation and external test set of the predictive models demonstrated a high-quality correlation between the observed and predicted toxicity values of all, training, test and validation sets in GA-ANN method. The model by ANN algorithm due to the lower error and higher regression coefficients was clearly superior to those models by MLR algorithm. The proposed model may be useful for predicting log LD50 of new compounds of similar class.