Analysis Of Chromatographic Data In Bio-aerosol Detection

Bio-aerosols are widely spread in the atmosphere.Not only may it threaten hu-man health,but also cause public security problems.In order to reduce the loss caused by bio-aerosols,with the development of research of bio-aerosol detection,a variety of bio-aerosol detection methods have been proposed.However,there is a lack of a real-time and specific method for bio-aerosol detection.In this thesis,a set of bio-aerosol detection equipment based on gas chromatography-mass spectrometry is built.After concentration,pyrolysis and methylation,chromatographic separation and mass spectrometry analysis,the chromatographic data are finally obtained.From chromato-graphic data,the chemical substances can be quickly identified in the sample,by which the species of microorganisms can be detected in the air.It is important to align peak for dealing with chromatographic data.We regard the problem of peak alignment of chromatographic data as a binary classification problem,and use Siamese Network to align chromatographic data.But the false positive rate of the model is higher in test datasets.In this thesis,we optimize the model,and reduce the false positive rate while maintaining the true positive rate.The main contents of this thesis are as follows:1)In order to detect the bio-aerosols in real time and identify its species,a mass spectrometer for detecting bio-aerosol was designed and built.The whole process,time sequence,gas circuit and interface of the instrument are designed and built.To begin with,the system concentrates air to obtain as many particles as possible;in addition,the particles are processed by pyrolysis and methylation to produce more volatile methyl-ester biomarker and separated by chromatography;thirdly,the ionization of CI source is used to electrify the derivatived biomarker,which can be detected by mass spectrometry;last but not the least,the algorithm of the spectrum is used to find out the possible components in the obtained aerosols.The whole process is less than half an hour.2)In order to realize automatic data analysis,the Siamese Network is used to align the chromatographic data.After thouroughly reviewing the existing algorithms of chro-matographic data alignment,the relevant machine learning theory and technology are combined.The alignment of chromatographic peaks is a problem of few-shot learn-ing.The model is constructed by using Siamese Network.The positive pairs and the negative pairs should be generated by original samples.With similarity learning,the classifier can classify learned class and unlearned class.3)For the structure of model,more than 50 ablation experiments are carried out to study the influence of each part of the network on performances of the model.And the best performance model with the different requirement is proposed.The experiment is designed to debug a series of structure and parameters of the network,which improves the average precision of identification.Aiming at the problem of peak alignment for bio-aerosol,Siamese Network is used to identify the peaks.Experiments show that the generalization ability of the model has been strengthened:in the test set,the model can not only be applied to the data categories that belong to the training set,but also be used for the data of those untrained classes,the performance of the model has been greatly improved(Average precision from 0.798 to 0.869).This is of great significance for better detection of bio-aerosol by chromatography.