MODELING OF HUMAN ACUTE TOXICITY FROM PHYSICOCHEMICAL PROPERTIES AND NON-VERTEBRATE ACUTE TOXICITY OF THE 38 ORGANIC-CHEMICALS OF THE MEIC PRIORITY LIST BY PLS REGRESSION AND NEURAL-NETWORK

Linear and non-linear modelling of human acute toxicity (as human leth al concentrations; HLCs) of the 38 organic chemicals from the 50 prior ity compounds of the Multicentre Evaluation of In Vitro Cytotoxicity ( MEIC) programme was investigated. The models obtained were derived eit her from a set of 23 physicochemical properties of the compounds or fr om their acute toxicities to five aquatic non-vertebrates together wit h the physicochemical properties. For the linear type, modelling was p erformed using a partial least square projection to latent structures (PLS) regression method; for the non-linear models, both PLS regressio n and neural network were utilized. A neural network using a combinati on of backpropagation and cascade-correlation algorithms was applied i n this study. The results generally reveal a slightly better predictiv e performance of the models obtained from PLS regression than those ob tained from neural networks. However, the model composed of physicoche mical properties (PC-model) from the trained neural network using a ba ck propagation algorithm with pruning technique proved superior to tha t trained with a combination of backpropagation and cascade-correlatio n algorithms after leave-one-out cross-validation. The predictive powe r of the PC-models, whether linear or non-linear, was comparable with that of the corresponding models consisting of both structural descrip tors and the ecotoxicological tests (ECOPC-models), except for the bat tery (ECOPC-model) from the neural network. The composition of the 'be st' PLS and neural network models points to the importance of the comb ination of physicochemical properties reflecting lipophilicity, size, volume, intermolecular binding forces and electronic properties of the molecule. All the aquatic non-vertebrate tests are shown to be essent ial in explaining human acute toxicity. However, the degree of contrib ution differed, with the crustacean (Arremia salina) and the bacterial (Microtox) bioassays being more important to the linear and non-linea r PLS models, whereas the crustacean (Artemia salina and Streprocephal us praboscideus) tests, and the rotifer (Brachionus calyciflorus) assa y were important to the neural network models. The organochlorine (lin dane) and bipyridinium (paraquat) pesticides were common outliers in a ll the models. Moreover, the latter two compounds and the organophosph ate (malathion) pesticide were also common outliers in all ECOPC-model s. Other types of pesticides, however, fit the models. The predicted H LCs of a number of non-pesticides, including some chlorinated compound s, also deviated from the observed HLCs by more than one order of magn itude.