Ecotoxicological classification risk index for soil (ECRIS) prediction of some phenolic compounds using multiple linear regressions and artificial neural network QSPR paradigm

Ecotoxicology is study of toxic chemicals effects on biological organisms, especially atpopulation, community, ecosystem and biosphere levels. Ecotoxicology is a multidisciplinary fieldintegrates ecology, toxicology, physiology, analytical chemistry, molecular biology, and mathematics.The ecotoxicological effects are changes in state or dynamics of an organism or at other levels ofbiological organization, resulting from exposure to a chemical which these levels may includesubcellular level, cellular level, tissues, individuals, populations, communities and finally ecosystems[۱-۲]. Ecotoxicological classification risk index for soil (ECRIS) is a new classification system specificfor soil risk assessment which gives a comparative presentation of risk linked to environmentalcontamination by any chemical [۳-۴]. Data set chemicals of QSPR modeling were found in variouslandfills leachate of north Italy which the ECRIS values ranged from ۱.۳۲ to ۵۸.۴۴ for ۲-Imidazolidinthyone and ۴.۴'-(Methylethylidene)bis-phenol. Linear and nonlinear models developedusing multiple linear regressions (MLR) and artificial neural network (ANN) approaches. The MLRmodel between six selected descriptors and desired ECRIS values was PECRIS =− ۹.۷۹۷ (± ۲.۹۴۸) + ۴.۶۱۹ * RDF۰۴۵v (± ۰.۶۷۶) + ۷.۳۵۱ * MLOGP (± ۱.۱۱۷) − ۱۰.۰۶۶ *∗Mor۱۳u (± ۲.۴۵۵) + ۶.۰۴۷ * HATS۴u (± ۲.۶۴۵) + ۵۶.۷۵۳ * HATS۵m (± ۲۰.۹۶۸) +۴.۲۹۹ * Hy (± ۱.۶۷۰)). In order to check any nonlinear relationships between selected molecularstructural descriptors and ECRIS values, artificial neural network was applied by using STATISTICAsoftware [۵]. A three-layer network with sigmoid transfer function was designed; ۶ descriptors andECRIS respectively were used as inputs and outputs values. After training of network, it was used topredict of ECRIS values of data set for training, internal and external test sets. Robustness and reliabilityof constructed MLR and ANN models were evaluated by using leave-one-out cross-validation methodthat respectively was equal Q۲MLR = ۰.۸۴ and Q۲ANN = ۰.۹۳.

کلیدواژه ها:

Ecotoxicology ، ECRIS ، QSPR ، Multiple linear regressions ، Artificial neural network