Quantitative structure-toxicity relationships of nitroaromatic compounds by linear and nonlinear chemometrics Methods

A quantitative structure–toxicity relationship (QSTR) study is suggested for the prediction of toxicity (IGC50) of nitroaromatics. Ab initio theory was used to calculate some quantum chemical descriptors including electrostatic potentials and local charges at each atom, etc. Among the investigation of QSAR, one of the most important factors affecting the quality of the model is the method to build the model. Many multivariate data analysis methods such as multiple linear regression (MLR) and partial least squares (PLS) have been used in QSAR studies [1]. MLR, as most commonly used, has been extensively applied to QSAR investigations. However, the practical usefulness of MLR in QSAR studies is rather limited, as it provides relatively poor accuracy. Modeling of the toxicity of nitroaromatic compounds as a function of molecular structures was established by means of the partial least squares (PLS) and least squares support vector machines (LS-SVM). The support vector machine (SVM) is a popular algorithm developed from the machine learning community. Due to its advantages and remarkable generalization performance over other methods, SVM has attracted attention and gained extensive applications. As a simplification of traditional of SVM, Suykens and Vandewalle have proposed the use of least-squares SVM (LS-SVM) [2]. LS-SVM encompasses similar advantages as SVM, but its additional advantage is that it requires solving a set of only linear equations (linear programming), which is much easier and computationally more simple.