Predicting Ecotoxicology of Organophosphorous Insecticides: Successful Parameter Selection with the Genetic Function Algorithm

Abstract

QSARs have been developed for 20 organophosphates, using good experimental data in the form of LC 50 toxicity values towards guppy from previously published work. A large number of steric, physical and electronic descriptors have been calculated. After the removal of descriptors that were highly cross-correlated or correlated poorly with activity, the genetic algorithm was employed to select the optimum descriptors for use in multivariate regression. Best equation r 2 and r 2 cv values were 0.943 and 0.905 respectively. Molecular ®eld Analysis was also used in conjunction with the Genetic Function Approximation (GFA) algorithm to obtain predictive equations, r 2 and r 2 cv values were 0.939 and 0.896 respectively. However, when the data were randomised, results proved to be as good, (and in one case better) than using the correct experimental values. For equations subsequently derived using the Genetic Partial Least Squares (GyPLS) technique (best r 2 and r 2 cv values being 0.943, 0.905 respectively), randomisation gave worse answers than using the correct experimental values. This work demonstrates the danger of using the powerful GFA technique to select from a large number of parameters, but establishes that the GyPLS method is suf®ciently robust to provide reliable results.

QSAR studies of multidentate nitrogen ligands used in lanthanide and actinide extraction processes

“…The QSAR module was then used within Cerius2 [24] to generate multiple regression equations. In previous work [27][28][29] we have found that the most useful method of variable selection is the Genetic Algorithm [30] which has the distinctive property that many where LSE is the least squares error (y obs − y calc ) 2 , c the number of descriptors plus one; p the number of independent parameters in the descriptors and M is the number of samples used in the training set. The advantage of using the LOF directly rather than the LSE is that the LOF does not decrease with increased number of descriptors and the lowest value is found for an equation with the optimum number of parameters.…”

Promises and Pitfalls of Quantitative Structure−Activity Relationship Approaches for Predicting Metabolism and Toxicity

“…An example can be found in a study describing QSAR models for the acute toxicity of organophosphorous compounds toward Daphnia magna and to the honeybee ( Apis mellifera ), using six descriptors to model a data set of 14 ( n / k = 2.3) and 22 ( n / k = 3.6) compounds, respectively . It should be realized that in general, the larger the number of descriptors used and the fewer the observations in the training set, the significantly higher the probability for the occurrence of a chance correlation ,, .…”

Panmictic and Clonal Evolution on a Single Patchy Resource Produces Polymorphic Foraging Guilds

“…In the early 1990’s, Stockwell, Nobel and colleagues developed machine-learning methods, in particular a framework called GARP (Genetic Algorithms for Ruleset Production) [ 13 ], which since then has been frequently used to study species distributions [ 14 – 16 ]. Genetic algorithms have also been used inter alia to study the evolution of female preference as it relates to male age [ 17 ], dispersal in insects [ 18 ] including the potential impact of climate change on the geographical distribution of the Argentine ants [ 19 ], insect-plant [ 20 ], and prey-predator studies [ 21 ], as well as ecotoxicology [ 22 ], landuse change [ 23 ] and other challenging questions [ 24 ].…”