Machine Learning-Based QSAR and Molecular Docking Analysis for Toxicity Prediction of Pesticides: A Tool for Environmental Risk Management

Ezadi, Fateme; Baher, Elham; Gharaghani, Sajad

Machine Learning-Based QSAR and Molecular Docking Analysis for Toxicity Prediction of Pesticides: A Tool for Environmental Risk Management

Articles in Press

Document Type : Original Article

Authors

Fateme Ezadi ¹

Elham Baher ¹

sajad gharaghani ²

¹ Department of chemistry, faculty of science, Golestan University, Gorgan, Iran

² Department of Bioinformatics, Laboratory of Chemoinformatics, Institute of Biochemistry and Biophysics, University of Tehran, Tehran, Iran

Abstract

In the present study, the lethal dose (LD50) of some pesticides was predicted using quantitative structure-activity (QSAR) studies. For this purpose, after drawing the molecular structure, molecular descriptors were calculated and LD50 by multiple linear regression (MLR) and support vector machine (SVM) was predicted. Genetic algorithm (GA) and MLR-stepwise were used to select the related descriptors. By comparing the obtained statistical results, the MLR-SVM model (Q2 =0.498 and SPRESS =0.047) was best model with R=0.99 and SE=0.014 and R=0.718 and SE=0.082 for the training and test set, respectively. Then, molecular docking was performed to investigate the ligand-protein interactions. For this part, the pesticides with the highest (LD50 = 2.68) and the lowest (LD50=2.21) toxicity were selected. Based on the obtained results, compounds that tend to form more hydrophobic bonds are causing more toxicity, and also the presence of some elements such as fluoride in their molecular structure increases their toxicity.

Keywords

QSAR studies

Molecular docking

Genetic Algorithm

Support vector machine

Subjects