Interactive effects of PAHs and heavy metal mixtures on oxidative stress in Chlorella sp. MM3 as determined by artificial neural network and genetic algorithm

Mixture toxicity studies are very complex due to the complexity exhibited by the chemicals involved, and the net interaction effects are highly dependent on mixture combinations, exposure dose and the test organism. For assessing the toxicity of mixtures, factorial analysis has been widely used, while the usage of models developed by artificial neural network (ANN) analysis and genetic algorithm (GA) is very limited. We combined for the first time the factorial design experiment with ANN and GA to develop a model for predicting the interactive toxicological effects using a soil microalga, Chlorella sp. MM3. The chemicals included in the mixtures were two polyaromatic hydrocarbons (PAHs), phenanthrene and benzo[a]pyrene, and two heavy metals (HMs), cadmium and lead. Three biochemicals implicated in oxidative stress, viz., malondialdehyde (a measure for lipid peroxidation, LPO), catalase activity and proline accumulation were used as the toxicity criteria. Validation of the predicted results related to the biochemicals with the experimental data clearly indicated that the model developed with the combination of ANN and GA is greatly effective in predicting the toxicity of PAHs and HMs mixtures toward microalga with < 10% relative error. Both catalase and LPO were found to be the promising biomarkers for predicting microalgal toxicity of PAHs and HMs mixtures. In addition, a significant positive correlation was evident between the removal of PAHs/uptake of HMs and LPO.