Modelling physico-chemical properties of halogenated benzenes: QSAR optimisation through variables selection

Abstract This paper describes the development of multivariate QSAR models for halogenated benzenes. Four physico-chemical properties have been modelled: Boiling Point (BP), Melting Point (MP), Flash Point (FP) and Density (D). For the multivariate characterisation of this class of compounds 90 topological indices derived from the chemical formula were used. The models were calibrated on training sets selected by a D-optimal design and were optimised by a statistical procedure of selection of the most informative variables. The predictive capacity of the resulting models measured in terms of the parameter SDEP (Standard Deviation of the Errors of Predictions) were as follows; MP: ± 21 ±C (with MP ranging from -48° to 181°C); BP: ±13°C (75° to 286°C); D: ±0.14 (1.02 to 2.52 g/cm(3)); FP: ±17°C (-12 to 126°C). These data were confirmed "experimentally", i.e., on external validation sets of compounds, the experimental values of the standard deviation of prediction errors were in fact as follows; MP: ±30°C; BP: ±16°C D: ±0.18 g/cm(3); FP: ±16°C. The developed models therefore provide a valuable tool for estimating physico-chemical properties relevant to environmental chemistry.