A series of quantitative structure-retention relationship models were developed to predict gas chromatographic relative retention times （GC-RRTs） for 209 polybrominated diphenyl ether （PBDE） congeners on 10 statio...A series of quantitative structure-retention relationship models were developed to predict gas chromatographic relative retention times （GC-RRTs） for 209 polybrominated diphenyl ether （PBDE） congeners on 10 stationary phases. A genetic algorithm with twofold leave-multiple-out cross validation （LMOCV） was used to select optimal subsets from large-size molecular descriptors. Overall multiple-linear regression fitting correlation coefficients （R2） are greater than 0.988, except for the CP-Sil 19 colunm, in which Q^uocv （correlation coefficient of LMOCV）, Q^oocv （correlation coefficient of leave-one-out cross validation, LOOCV）, and Rp2re （correlation coefficients of prediction set） are larger than 0.98. The excellent statistical parameters reveal that the models are robust and have high internal and external predictive capability. According to the descriptors for constructing the models, the GC-RRTs in various stationary phases are highly dependent on distances among atoms, branches of molecules, and molecular properties. PBDE congeners with 1, 9, and 10 bromines are major outliers based on the results of the application domain.展开更多
基金Project supported by the Guangxi Natural Science Foundation （No. 2011GXNSFA018061）, the Scientific Research Fund of Guangxi Education Department （No. 200708LX265）, the National Nature Foundation Committee of China （No. 21167006）, and 863 Advanced Research Project （No. 2007AA06Z416）.
文摘A series of quantitative structure-retention relationship models were developed to predict gas chromatographic relative retention times （GC-RRTs） for 209 polybrominated diphenyl ether （PBDE） congeners on 10 stationary phases. A genetic algorithm with twofold leave-multiple-out cross validation （LMOCV） was used to select optimal subsets from large-size molecular descriptors. Overall multiple-linear regression fitting correlation coefficients （R2） are greater than 0.988, except for the CP-Sil 19 colunm, in which Q^uocv （correlation coefficient of LMOCV）, Q^oocv （correlation coefficient of leave-one-out cross validation, LOOCV）, and Rp2re （correlation coefficients of prediction set） are larger than 0.98. The excellent statistical parameters reveal that the models are robust and have high internal and external predictive capability. According to the descriptors for constructing the models, the GC-RRTs in various stationary phases are highly dependent on distances among atoms, branches of molecules, and molecular properties. PBDE congeners with 1, 9, and 10 bromines are major outliers based on the results of the application domain.
