Research On Separation Of Microbial Single Cells Based On Raman Optical Tweezers Technology And Deep Learning

The basic unit of living things is the cell.It is at the single-cell level that research can be conducted to fully understand the nature and laws of life activities.However,the majority of microorganisms change their biochemical characteristics after being removed from the in situ environment,leading scientists to be unable to do further studies on microorganisms.Raman optical tweezers technology is considered as a noninvasive method in single-cell analysis,which can study microorganisms in situ environment.Raman spectroscopy is known as a biological fingerprint and can provide biochemical information about a bacterial sample.Raman spectra contain a large number of spectral features,and traditional Raman spectral analysis methods use manual feature extraction with limited expression and weak generalization of spectral features,which cannot meet the demand for higher accuracy in classification tasks.The deep learning model is similar to the processing of human visual system with a hierarchical structure,which can automatically extract the high-level semantic information of Raman spectra.Although great progress has been made in the research of Raman spectra classification using deep learning method,there are still great challenges.For example,the sample data of Raman spectrum is too small,and the data characteristics of Raman spectrum cannot be effectively utilized.Therefore,this paper mainly studies the separation of microbial single cells by Raman optical tweezers technology based on deep learning.Through literature reading,this paper summarizes the relevant research progress in this field,and studies and explores the relevant technical essentials in this field.In order to realize the analysis and separation of in situ microorganisms by combining Raman optical tweezers technology and deep learning,a Raman optical tweezers-microfluidic separation platform was built,and two microfluidic chip designs for Raman optical tweezers separation were proposed,which met the needs of in situ research of single-cell microorganisms.In order to solve the problem that the amount of Raman spectral data is small and it is difficult to establish a database,a spectral data generation and classification method based on the combination of PGGAN-Res Net network and Raman optical tweezers technology is proposed.The database can be established with a small amount of data,and has been successfully applied to the identification of microorganisms in deep-sea garbage;A method for generating and classifying Raman spectral data based on VAE-LSTM network is proposed,which reduces the number of Raman spectral data collected,and has been successfully applied to the identification and identification of clinical pathogenic bacteria;In order to solve the problem that the characteristics of Raman spectral data can not be effectively used,a transformer-based classification method of Raman spectral data is proposed,which makes full use of the characteristics of Raman spectrum,and has been successfully applied to the recognition research of cold seep bacteria in the deep sea.The proposed method has achieved excellent results in different classification tasks.The main research contents of the article are as follows:(1)Explore the principles of optical tweezers technology and Raman spectroscopy technology,and elaborate on the construction of a Raman optical tweezers microfluidic sorting platform.In harsh environments(such as deep-sea environments),a microfluidic sorting chip with a partial pressure structure was proposed to meet the stability analysis of microorganisms using Raman optical tweezers in situ environment;A microfluidic sorting chip based on oil in water principle is proposed to avoid contamination of cells sorted in situ.To lay the corresponding technical foundation for subsequent related research.(2)In order to reduce the problem of difficulty in collecting a large number of Raman spectra in harsh environments,a study was conducted on the Raman spectral classification of microorganisms in deep-sea garbage based on progressive generative adversarial networks and residual networks.A Raman spectral classification method based on progressive generative adversarial networks and residual networks was designed,which expanded the Raman dataset and reduced the number of Raman spectra collected.Compared to traditional deep learning algorithms,this algorithm has higher accuracy and robustness.(3)In order to solve the problems of difficult spectral collection and poor spectral quality,a study was conducted on the Raman spectral classification method of microorganisms based on variational autoencoders and short-term memory neural networks.A microbial Raman spectrum classification method based on the combination of a variational self encoder and a short-term memory neural network was designed.This method can not only increase the number of samples in Raman spectroscopy,but also improve the accuracy of classification.A detailed analysis was conducted on the differences between Raman spectra generated by variational autoencoders and real spectra.Compared with the current advanced Raman spectroscopy classification methods,the proposed method has reached the current leading level.In order to verify the applicability and stability of the proposed method,samples were used for testing to classify the spectral data of untrained strains,with an average recognition accuracy of up to 95%.(4)In order to fully utilize Raman spectral information,a study was conducted on the Raman spectral classification method of deep-sea cold spring microorganisms based on Transformer.A Transformer classification method was designed for the Raman spectra of deep-sea cold spring bacteria.Fully utilizing the data features of Raman spectroscopy,this method has higher accuracy compared to traditional deep learning algorithms.