QSPR study of polychlorinated diphenyl ethers by molecular electronegativity distance vector (MEDV-4)

Polychlorinated diphenyl ethers (PCDEs) have received more and more concerns as a group of ubiquitous potential persistent organic pollutants (POPs). By using molecular electronegativity distance vector (MEDV-4), multiple linear regression (MLR) models are developed for sub-cooled liquid vapor pressures (PL), n-octanol/water partition coefficients (KOW) and sub-cooled liquid water solubilities (SW,L) of 209 PCDEs and diphenyl ether. The correlation coefficients (R) and the leave-one-out cross-validation (LOO) correlation coefficients (RCV) of all the 6-descriptor models for log PL, log KOW and log SW,L are more than 0.98. By using stepwise multiple regression (SMR), the descriptors are selected and the resulting models are 5-descriptor model for log PL, 4-descriptor model for log KOW, and 6-descriptor model for log SW,L, respectively. All these models exhibit excellent estimate capabilities for internal sample set and good predictive capabilities for external samples set. The consistency between observed and estimated/predicted values for log PL is the best (R = 0.996, RCV = 0.996), followed by log KOW (R = 0.992, RCV = 0.992) and log SW,L (R = 0.983, RCV = 0.980). By using MEDV-4 descriptors, the QSPR models can be used for prediction and the model predictions can hence extend the current database of experimental values.