–ß‚é
JP28

A novel prediction model for aqueous solubility of polychlorinated biphenyl isomers based on their molecular topological characteristics.

›Masakazu MakinoiGraduate School of Nutritional and Environmental Sciences, University of Shizuokaj

The use of quantitative structure - property relationship to predict the logarithmic aqueous solubility coefficients, logSw, of 136 congeners of polychlorinated biphenyls, PCBs, known as environmental contaminants is reported. Principal component regression, PCR, and computational neural networks, CNNs, were used to develop models using a set of information-rich descriptors, which were encoded information about molecular topology, geometry, and net charge at atoms calculated by using MOPAC 93 rev. 2 software package. We compared the square of the regression coefficient, r2, values obtained from two models. For the congeners, both the models seem to be appropriate for the prediction, since the r2 values were 0.953(PCR) and 0.982(CNNs), respectively. However, it was found that they were insufficient for predicting logSw of chlorinated biphenyl, CB, isomers. Therefore, we reconstructed prediction models for the CB isomers by using them. It was elucidated that CNNs derived! adequate models for predicting logSw of CB isomers, and that they were markedly preferable to the models derived from PCR judging from the r2 values.