A QSTR Study Of The Acute Toxicity Of Phenol Derivatives To Aquatic Tetrahymena Pyriformis

At present, with the continuous development of synthesis industry,phenol derivatives are widely used as chemical raw materials.Because of its obvious biological toxicity, endocrine disruption andbiological accumulation, phenol derivatives become ubiquitouspollutants that has yield adverse effects on humans and other livingspecies. So these xenobiotics have inspired a wealth of experimentalstudies for assessing their ecological risk. Subsequent to theselaboratory investigations, numerous quantitative structure–toxicityrelationship (QSTR) models have been derived on these chemicals tosave time and money, to understand the mechanisms underlying theirtoxicity and to simulate the ecotoxicological behavior ofnon-synthesized compounds, as well as to satisfy public resistance toanimal testing. And the QSTR studies have made a lot ofachievements in the field of environmental science. It is necessary to establish the QSTR models of the toxicity of phenol derivatives.Tetrahymena pyriformis toxicity test method is developed in thelate1990s. It is a new type of biological toxicity test method and hasmany characteristics, such as quick detection, simplicity, economicalefficiency, wide applications, etc. So it is widely used in medicine,organic, inorganic, and toxicological evaluation of water pollutant.258toxicity data of phenol derivatives to aquatic Tetrahymenapyriformis were selected from literature in this dissertation, and themodels were established using7molecular descriptors. In the QSTRstudies, multiple linear regression, partial least squares regression,and BP neural network were used respectively to solve the problems.The main work of this dissertation:(1)The chemometric methods applied in the dissertation weredescribed briefly, which include multiple linear regression, partialleast squares regression, BP neural network and principal componentanalysis.(2)The molecular descriptors of258phenol derivatives werecalculated and selected by ADMEWORKS ModelBuilder software(Version4.5Standard), and finally selected7descriptors as variablesto develop models.(3)A robust diagnostic method was used to eliminate24outliers,and then the sphere exclusion algorithm was applied to divide the whole date set into training and internal test sets rationally. It wasrequired the internal test set and the independent external validationset were uniformly distributed into an overlapping area of the PCAspace. Finally, it was divided into3reasonable date set.(4)Combined with the toxicity data selected from the literature,multiple linear regression, partial least squares regression, and BPneural network method were applied respectively to establish theQSTR models. The toxicity prediction models were successfullyestablished. Then for the external validation set, consensus modelingmethod was used to yield more accurate predictions.According to the QSTR research results, following conclusionswere obtained: the models show their high precision and goodprediction abilities. Compared with MLR and PLS models, BP neuralnetwork models have their advantages. That is to say, nonlinearmodels are better than the linear ones for the performance. However,BP neural network models can not give the mathematical formuladirectly. Moreover, MLR and PLS models are more simple. We onlyneed to know the molecular structures of phenol derivatives, and usethe QSTR prediction models to predict the toxicity values reasonably.This method shows its potential abilities on the rapid identification oftoxicity of phenol compounds, as well as the synthesis of phenoliccompounds with low toxicity.