Prediction of toxicity and octanol – water partition coefficient of organochlorine pesticides using Molecular Descriptors and GA-MLR Method

Subject Areas : entomology and othea arthropods

F. Shafiei ¹ , Z. Zamani ²

1 - Department of Chemistry, Arak Branch, Islamic Azad University, P.O. BOX 38135-567, Arak, Iran
2 - PHD student, Department of Chemistry, Arak Branch, Islamic Azad University, P.O. BOX 38135-567, Arak, Iran

Received: 2019-07-07 Accepted : 2019-11-25 Published : 2019-11-22

Keywords:

Abstract :

In this research, Quantitative Structure–Activity Relationship (QSAR) studies have been used to predict activities of organochlorine pesticides. Firstly, the chemical structure of molecules was drawn with the Gauss view 05 program and optimized at Hartree–Fock level of theory and 6-31G* basis sets using Gaussian 09 software. The physiochemical properties namely octanol-water partition coefficient (logP) and toxicity (log LD50) are taken from the scientific web book. The dragon software has been used for the calculation of molecular descriptors. The suitable descriptors were selected with the aid of the genetic algorithm (GA) and backward techniques. At the next step, the relationship between molecular descriptors and the activities was investigated by multiple linear regression (MLR) method. In order to build and test QSAR models, a data set of organochlorine pesticides was randomly separated into 2 groups: training (80%) and test (20%) sets. The models were evaluated with regression parameters: correlation coefficient (R), squared regression coefficient (R2), adjusted correlation coefficient (R2 adj) and root mean squared error (RMSE). For the predictive ability and verification of the models are discussed by using Leave-One-Out (LOO) cross-validation and external test set. The external prediction accuracy of the obtained models was examined using the above regression parameters. Results of validations and high statistical quality of models indicate that generated GA-MLR models are reasonable QSAR models. These models help to delineate the important descriptors responsible for predicting their activities.