Application of the replacement method as novel variable selection in QSPR. 2. Soil sorption coefficients

We predict the soil sorption coefficients of 163 non-ionic organic pesticides performing a QSPR treatment. A pool containing 1247 theoretical descriptors is explored simultaneously encoding different aspects of the topological, geometrical, and electronic molecular structure. The application of Forward Stepwise Regression, Genetic Algorithms and the Replacement Method leads to an optimal six-parameter equation characterized with R = 0.949 and that also exhibits good cross-validated predictive ability, Rl-25%-o = 0.916. This model compares fairly well with a previously reported QSPR on the same data set with R = 0.904.