Research On Proteome Analysis Method Based On Convolutional Neural Network

Proteome research is crucial to a comprehensive understanding of the law of biological information.The main goal of proteome analysis is to identify and quantify proteins in experimental samples.The data-independent acquisition method based on tandem mass spectrometry is a main way to obtain proteomic data.The collected mixed MS2(Second stage of Mass Spectrometry)is composed of multiple peptide fragments simultaneously fragmented,which increases the complexity of peptide identification and quantification.At present,the main qualitative and quantitative research work needs to rely on chromatographic information,which will be affected in different chromatographic gradient data and can not be applied to the data without chromatography.Therefore,this thesis proposes a method based on Convolutional neural network(CNN)that does not rely on chromatographic information for qualitative and quantitative analysis of the data with chromatography and without chromatography.The main work of this thesis includes:(1)At present,the qualitative method of spectrum matching similarity requires manual design of scoring functions,and the method of constructing chromatographic peaks needs to extract a large number of chromatography-related features,which will be affected by ion co-elution and chromatographic dimensions.Aiming at the qualitative problems,a qualitative model based on CNN is proposed,which does not need to extract features related to chromatographic peaks,and uses the network to score peptides automatically.Performing experiments on mixed samples and samples of different concentrations improves qualitative reproducibility and reliability.(2)Aiming at the problems that peptide quantitative methods require complex processing procedures,rely heavily on chromatographic information,and are affected by chromatographic dimensions,a quantitative model based on CNN is proposed,which does not need to construct chromatographic peaks and directly obtain the quantitative value of peptide through regression prediction.Experiments on mixed samples and complex plasma samples validated the accuracy and generalization of the quantitative model.(3)At present,the research relying on chromatographic information can not be applied to the data without chromatography.Only a few studies use modified spectrum matching similarity tools for qualitative analysis,but the effect is not good,and the label-free data can not be analyzed quantitatively.To solve these problems,this thesis migrates the previously proposed model based on CNN to the data without chromatography.For qualitative,the model was directly transferred to the processed data without chromatography or retrained to obtain a qualitative model for the data,and the test was carried out on repeated experiments to verify the effect of model migration.For quantification,the previously trained model was transferred to isotope-labeled and label-free data.The results show that the model can accurately quantify the data without chromatography.