Development of a 3D QSPR model for adsorption of aromatic compounds by carbon nanotubes: comparison of multiple linear regression, artificial neural network and support vector machine

Abstract

Adsorption coefficients of 39 aromatic compounds onto multi-walled carbon nanotubes have been compiled. To understand the relationship between adsorption coefficients and physicochemical properties of aromatic compounds, a 3D quantitative structure–property relationship (QSPR) model was developed by the utilization of 3D molecular structures of 39 aromatic compounds. A Monte Carlo computational algorithm was utilized to generate 3D molecular descriptors and physicochemical properties for the QSPR model. Of the physicochemical descriptors: log K_ow, number of nitrogen and oxygen atoms and number of atoms in rings present positive correlation. However, the dipole moment of the molecule and number of hydrogen bonds accepted by the solute present negative correlations. In the model development process, three different learning approaches, multiple linear regression (MLR), artificial neural network (ANN) and support vector machine (SVM), were used. The validation results showed that SVM- and ANN-based models resulted in a better agreement between predicted and measured values, with the coefficient of determination (R²) of 0.8317 and 0.7829, than the MLR-based model with R² of 0.5093.