Modeling and quantitative structure-property relationship studying to predict the half-life of polychlorinated biphenyls using multivariate linear regression and artificial neural networks

Document Type : Original Article

Authors

• S B ¹
• M N ²
• A T ²

¹ Shahid Beheshti University

² IAU

Abstract

Quantitative structure-property relationship (QSPR) study was performed to predict the half-life of some polychlorinated biphenyl derivatives using multivariate linear regression (MLR) and artificial neural networks (ANN). First, the structure of the compounds, the drawing, and the appropriate group of descriptors were calculated. Then, the step-wise method was used to obtain the best descriptors that were most related to the half-life of the compounds. With this method, 6 descriptors including Lop, GATS5m, GATS8m, LDip, RDF020u, R2v + were selected from the types of topological descriptors, charge, three-dimensional representation of molecules based on electron diffraction and radial distribution function. First, a multiple linear regression linear model was constructed. Then, artificial neural network was used to obtain better results. The values of coefficient of determination (R²) and root mean square error (RMSE) for the test series were equal to 0.716 and 0.050 for the MLR linear model and 0.896 and 0.030 for the nonlinear ANN model, respectively. Statistical data show the superiority of ANN method over MLR method.

Keywords

• Quantitative structure-property relationship
• polychlorinated biphenyls
• half-life
• multiple linear regression
• artificial neural network

---