Research on Neural Network Theory of the Toxicity of Aromatic Hydrocarbons on <i>Chlorella vulgaris</i>

Abstract

Using BP neural network method, we calculate and analyze the molecular structure of aromatic hydrocarbons. Then, we get the electrotopological state indices and the molecular electronegativity distance vectors of 25 aromatic hydrocarbons based on the calculation of molecular structure characteristics and adjacency matrix. By regression, we get and optimize the structural parametersE9,E13,E17andM15. The four structural parameters are used as the input variables and a 4-2-1 network structure is employed to construct a BP artificial neural network model for predicting acute toxicitypEC50. The total correlation coefficientRis 0.994 and the average error between the predicted value and experimental value ofpEC50is 0.079, which indicate that the ANN model has good stability and superior predictive ability. The results show that there is a good nonlinear correlation between acute toxicitypEC50and the four structural parameters. The results of our research reveal that the toxicity of aromatic hydrocarbons is closely affected by electrotopological state indices and the molecular electronegativity distance vectors. Therefore, it will be helpful in assessing the hazard of aromatic hydrocarbons to environment.