Study On QSAR Model Of Acute Toxicity Of Fathead-minnow Based On Improved Elman NN

Organic compounds have been widely used in industrial production and life fields,resulting in a large amount of organic residues in the environment.These organic compounds migrate and transform through air,water,soil,sediments and other environmental media,posing a potential threat to the health of animals,plants and humans in the ecosystem.At present,the total number of registered and synthesized chemical substances has exceeded 100 million,and only 300,000 are under supervision.Therefore,reasonable supervision and use of the toxicity information obtained from these huge chemical substances is the fundamental method to protect human health and the ecological environment from harm.However,most of its compounds lack information on their toxicity,and there are huge gaps in the data.Traditional animal experiment methods are time-consuming,labor-intensive,expensive,and timeconsuming,and violate the 3R principles of animal ethics.In order to fill the gaps in the toxicity data of organic compounds efficiently and quickly,quantitative structure-activity relationship(QSAR)has emerged as an important alternative method.This method uses the molecular structure of organic matter as a variable and establishes a calculation model to predict the physical,chemical or biological activity of the organic matter,thereby revealing the structural factors that affect certain properties of the compound,and understanding and predicting the onlooker structure of the substance at the molecular level The impact on a certain characteristic.At present,this method is widely used in environmental chemistry,medicinal chemistry,material chemistry and other chemistry-related fields.The blackhead fish is a very important biological model in the aquatic environment toxicology research,because it is a kind of organism with the highest nutritional level in the aquatic food chain.Therefore,this paper starts from the two aspects of the standardization of QSAR model modeling and the improvement of model performance,to study new methods of QSAR modeling and apply them to the research of drugs,environment and material chemistry to solve actual production problems.In this paper,taking the acute toxicity data of blackheaded fish as the research object,a new type of Cellular Automata-based Coyote Optimization Algorithm(WCCOA)is proposed to optimize and train the QSAR model of the improved Elman neural network.First,in view of the high-dimensional redundancy of the molecular descriptors of organic compounds,a local linear embedding algorithm with mixed distance(MSLLE)is used to reduce the dimensionality of the auxiliary variable sample data.The Euclidean distance is replaced by the homogenized distance and the Mahalanobis distance to improve the similarity between samples,and calculate the distance between samples to make the distribution area of the samples uniform between samples.This method can not only maintain strict globality to the greatest extent,but also take into account the local mapping ability.Secondly,an Elman neural network model with a feedforward adjustment strategy is proposed,which uses a receiving layer with memory function to dynamically adjust the hidden layer and output layer,and the nonlinear processing capability of the model is enhanced.Subsequently,the COA algorithm was improved by integrating Levy flight mechanism,adaptive weight factor and topology structure.Experimental results show that the improved WCCOA algorithm can effectively balance the diversity and convergence of the population,and improve the search accuracy and search speed.Finally,using the coefficient of fitting determination of the training set,the prediction accuracy index of the test set and the mean square value of the prediction error as the evaluation indicators,the robustness and predictive ability of the QSAR model constructed for the acute toxicity data of the blackheaded fish were tested through the external independent sample test set.Comprehensive experimental results show that the proposed QSAR model has good prediction accuracy and reliability.