Construction Of Models For Identification Of Intestinal Microorganisms In Urechis Unicinctus Based On Raman Characteristics

The efficient identification of intestinal microorganisms is of great significance in the application of basic research under the background of increasingly urgent biosafety requirements.Existing identification methods mainly include traditional culturable methods and molecular biology based culture free methods,but the efficiency of either method is relatively modest.Raman feature-based identification methods have been widely used and developed in recent years,and the introduction of deep learning into the analysis of Raman features is helpful to further improve the identification efficiency.This study begins with the isolation of gut microbes from marine organisms,expands the Raman database of gut microbes,and presents a preliminary experimental analysis focused on gut microbes,and finally constructs a more optimized model for the efficient identification of gut microbes.Urechis unicinctus is the only species of Urechis tubulosa in coastal areas of China,which has high economic value.Due to overfishing,the yield of Urechis unicinctus is decreasing year by year,which leads to the fact that the majority of Urechis unicinctus in the current market comes from aquaculture.However,the diseases of Urechis unicinctus occur frequently,especially the diseases caused by the imbalance of intestinal flora.Based on this,this study explored the intestinal flora of Urechis unicinctus,and obtained 8 strains by separation and purification;By means of Gram staining and analysis of the optimum growth conditions,a strain of bacteria was studied.It was found that the strain was gram negative,and the optimum growth temperature was 30 ?,p H value was 6.0 and salinity was 35 ‰;The results showed that the strain was resistant to many kinds of antibiotics and highly sensitive to chloramphenicol,carbenicillin,ofloxacin and ceftazidime.Then,based on the above eight bacteria,the existing Raman identification models were optimized to solve the problems of low classification accuracy,long training time,large amount of training data and unclear classification basis.The longterm short-term memory network(LSTM)improves the classification accuracy of the model through learning the correlation between the nodes before and after.The results show that the final classification accuracy of the model reaches 98%;The model based on the combination of autoencoder and quantum neural network(QNN)can speed up the data calculation in high-dimensional space and reduce the training time.The results show that the final classification accuracy of the model reaches 82%;By generating new data to assist training,GAN reduces the amount of data required for model training.The results show that the accuracy of the model based on small samples reaches 96%;Finally,the model based on self-designed loss function and classification label can limit the information of model selection and learning,and achieve the classification effect based on nucleic acid information.By testing the sensitivity of the model,the region associated with nucleic acid information in Raman data is preliminarily identified,and the results show that the region is associated with adenine of bacteria.The four deep learning models constructed in this study improve the classification effect of the original convolutional neural network;In particular,the construction of GAN network provides the possibility for model training based on small sample datasets,and is expected to provide a theoretical basis for the efficient identification of intestinal microorganisms of Urechis unicinctus.