Ranking-based selection of non-linear quantitative structure-property relationship models for prediction of bioconcentration factor of triazine derivatives as pesticide candidates

Abstract

The estimation of ecotoxicity and bioaccumulation of compounds as pesticide candidates is an important step in the estimation of their potential practical use. The present study is aimed to form several non-linear regression models based on artificial neural networks (ANN) for prediction of bioconcentration factor of a series of 6-chloro-1,3,5-triazine derivatives and to their ranking and selection based on sum of ranking differences (SRD) approach. The obtained networks represent quantitative structure-property relationship (QSPR) models. The input variables were selected based on hierarchical forward selection procedure and those are the following molecular descriptors: ATSm5 (autocorrelation descriptor mass descriptor weighted by scaled atomic mass), minHBa (minimum E-states for (strong) hydrogen bond acceptors), sumI (sum of the intrinsic state values) and DELS2 (sum of all atoms intrinsic state differences, measure of total charge transfer in the molecule). The total number of the established QSPR models was twelve and all models were validated and confirmed to be of high statistical quality and significant predictive ability. In order to rank and select the most suitable networks, the SRD approach was applied based on row average as the reference ranking.

Keywords

• bioaccumulation
• chemometrics
• ecotoxicity
• triazines
• qspr