Application of the characterisitc root index model tothe estimation of physico-chemical and biological properties of selected endocrine disrupting chemicals

Abstract

Quantitative structure-property relationship (QSPR) models based on two differentsets of parameters have been developed for water solubility (S), n-octanol/water partitioncoefficient (KOW), and Henry̕s Law Constant (H) of selected endocrine disrupting chemicals (EDCs). Similarly, QSPR models have also been established for the biologicalproperties including 50% effective inhibition concentration (48h-EC50) to algae andbioconcentration factor (BCF) in fish for substituted benzenes having endocrine disruptingproperties. The parameters used to develop these QSPR models were topology based Characteristic Root Index (CRI) and semi-empirical molecular descriptors, namely =energies of the highest occupied and the lowest unoccupied molecular orbital (EHOMO andELUMO), and dipole moment (u).The best fit equation found by "forward multiple linear regression" showed that the topology based CRI was the most important parameter for the modeling of S, KOW andBCF. For KOW a two-parameter equation including the CRI and EHOMO was obtained with acorrelation coefficient of r = 0.992, whereas a three-parameter equation including the CRI,ELUMO and u was obtained for both solubility and Henry̕s Law Constant with a correlation coefficient of r = 0.986 and r = 0.933, respectively. EHOMO and u didn̕t appear in the samemodel because of the collinearity. The CRI and ELUMO descriptors were used in regressionsof 48h-EC50 with a correlation coefficient of r = 0.926 while only the CRI had appeared asa descriptor for the prediction of BCF with a correlation coefficient of r = 0.850. Since the number of BCF data is limited to 18 for the CRI-based QSPR model (mod el 6), reliabilityof this model was first checked by classical tests, and then it was compared with the otherCRI-based model (model 7) reported in the literature. The results of modified jackknifetests indicated that the five models were statistically robust. Mean deviation of calculatedvalues from experimental data amounted to 0.27, 0.17, 0.28, 0.19 and 0.29 log units for theS, KOW, H, 48h-EC50, and BCF, respectively. The developed models have been used topredict the S, KOW, H, 48h-EC50, and BCF of those compounds where there are noexperimental measurements.