A new molecular representation, molecular hologram, is employed to investigate the quantitative rela- tionships between gas chromatographic retention indices (GC-RI) and molecular structures of polychlorinated diben- ...A new molecular representation, molecular hologram, is employed to investigate the quantitative rela- tionships between gas chromatographic retention indices (GC-RI) and molecular structures of polychlorinated diben- zofurans (PCDFs). Together with application of partial least squares (PLS) regression, the quantitative structure-reten- tion relationship (QSRR) model is constructed for GC-RI of 135 PCDFs. This new QSRR model presents high statistical quality and predictive value with crossvalidation correlation coefficient q2LOO values of 0.998, and non-crossvalidation correlation coefficient r2 of 0.998. 100 PCDFs are selected randomly as training set and the rest as testing set. The re- sult of PLS regressive analysis of training set yields r2 of 0.998 and q2LOO of 0.997. The GC-RIs of testing set are pre- dicted, and the correlation equation indicates that the model based on training set has excellent ability to predict the GC-RIs of PCDFs in testing set.展开更多
基金This work was supported by the National Basic Research Program of China(Grant No.2003CB4 l5001)the National Natural Science Foundation of China(Grant No.20477018)+1 种基金European Commission International Scientific Cooperation Project(Grant No.ICA4-CT-2001-10039)the Doctoral Program of Higher Education of China and the Core University Program.
文摘A new molecular representation, molecular hologram, is employed to investigate the quantitative rela- tionships between gas chromatographic retention indices (GC-RI) and molecular structures of polychlorinated diben- zofurans (PCDFs). Together with application of partial least squares (PLS) regression, the quantitative structure-reten- tion relationship (QSRR) model is constructed for GC-RI of 135 PCDFs. This new QSRR model presents high statistical quality and predictive value with crossvalidation correlation coefficient q2LOO values of 0.998, and non-crossvalidation correlation coefficient r2 of 0.998. 100 PCDFs are selected randomly as training set and the rest as testing set. The re- sult of PLS regressive analysis of training set yields r2 of 0.998 and q2LOO of 0.997. The GC-RIs of testing set are pre- dicted, and the correlation equation indicates that the model based on training set has excellent ability to predict the GC-RIs of PCDFs in testing set.
