Quantitative structure-activity relationship models for bee toxicity

Quantitative structure-activity relationship models for bee toxicity have been built up using CORAL software (http://www.insilico.eu/coral). The approach is based on the Monte Carlo technique. The molecular structure for the quantitative structure-activity relationship analysis has been presented by the simplified molecular input-line entry system. The so-called balance of correlations and the traditional scheme of building up quantitative structure-activity relationship models are compared in this work. The traditional scheme is based on three basic sets of compounds: training, calibration, and validation, whereas, the balance of correlations is based on four sets: active training, invisible training, calibration, and validation. As rule, the balance of correlations gives better models in comparison with the traditional scheme. The statistical characteristics of the models are quite good. Possible mechanistic interpretations and indications for the domain of applicability of these models are suggested.