Prediction of the Daphnia acute toxicity from heterogeneous data

Two descriptors (log(P(ow)), 'hardness') were selected to predict the Daphnia acute toxicity of a training set of heterogeneous chemical compounds. The data were extracted from 523 notification files about new chemicals stored at the French Department of Environment. The selection of the descriptors was carried out using a statistical method coupling ordinary least square (OLS) regression and genetic algorithm (GA). The validity limits for the final equation are discussed by comparing the actual and predicted activities of several compounds. The study points out the interest of the 'hardness' parameter for quantitative structure-activity relationships (QSAR) with a heterogeneous data set.