3D-QSPR Study For New Indices Defined Based On The Spatial Coordinates And Physicochemical Properties Of Polychlorinated Biphenyls

Quantitative structure-property relationship (QSPR) has become an important branch of chemometrics, can be applied to predict unknown property of compounds. It proves to be the significant measure in ecological risk evaluation of organic pollutants for its predicting ability for the migration, transformation and distribution of organic pollutants in the environment.The main objective of the paper is to study3D-QSPR of Polychlorinated biphenyls (PCBs), which are a group of important persistent man-made organic pollutants, they have some characters which related to the physicochemical properties, such as environmental persistence, bioaccumulation, long distance migration ability and so on, these physicochemical properties could not get from experiment, and therefore, we need to define or select descriptors to establish models to predict them. Theoretically, PCBs have209kinds of congeners, up to now,150kinds of which have been identified and most of which are not planar compounds, so their three-dimensional structure information may be used to predict their chromatographic and physicochemical properties.With the application of ChemWindow software, we drew molecular graphs of209PCBs, got their3D-structure calculated by SymApps6.0, and created molecular spatial coordinates matrix M for every3D-structure corresponds to its coordinates. We defined three indices, YS, YF and YW, YS was average value of the distance of all the atoms to the origin in the matrix M, a novel spatial distance index YF which calculated by the frobenius norm of M of PCBs, molecular spatial distance index YW was defined based on the Wiener index. According to a contrast and results analysis of linear regressions and BP Neural Network, QSPR models were established to estimate and predict chromatographic and physicochemical properties of PCBs with these indices, the predictive error of BP Neural Network is smaller than linear regression’s. It shows that BP Neural Network can give relatively excellent prediction to target with its nonlinear mapping ability comparison with linear regression. Besides, cluster analysis method based on the defined topological indices was used to classify the different numbers of chlorine atom in PCBs.In brief, the methodology developed in this paper can be useful to estimate physicochemical properties, and also provide a new idea and method for chromatographic analysis and QSPR research.