Modeling Of Peptide Fragment Ion Intensities Based On Deep Learning

Since its emergence,tandem mass spectrometry has become one of the most important technologies in proteomics.Peptide fragments in tandem mass spectrometer contain several main ion types,such as regular ions,internal ions,immonium ions and neutral losses.The masscharge ratios of these ions can be calculated accurately according to the simple rules of peptide breaking reaction.However,the ionic(relative)intensity cannot be predicted arithmetically from the amino acid sequence of the peptide.Predicting the intensity of these ions and understanding the fragmentation process are very important for protein identification.In proteomics based on tandem mass spectrometry(MS/MS),the prediction of theoretical mass spectrometry is of great significance for the identification of peptide sequences.In recent years,some researchers have devoted themselves to the prediction of theoretical mass spectra of polypeptides,including dynamic model-based methods and machine learning-based methods.However,these methods have some shortcomings.In this paper,a prediction model of ionic intensity of peptide fragments based on sequence to sequence(seq2seq)is proposed.Compared with other models used to predict the ionic strength of fragments,pep2 peaks can not only predict the intensity of regular ions,but also predict the intensity of internal ions.At the same time,the input of pep2 peaks is not limited by the length of the peptide sequence,and the output is not limited by the number of charged ions.In this paper,we validate the predictive performance of pep2 peaks for regular ions over 0.95 median PCC,for internal ions over 0.90 median PCC,and for pep2 peaks with good generalization and anti-jamming ability through interaction experiments between multiple sets of data.In addition,the data characteristics of internal ions are analyzed in detail.In the data processing part,the reliability of 20 ppm labeling error and the possibility of internal ions with different lengths in secondary mass spectrometry are demonstrated.In the experimental part,pep2 peaks were used to research the internal ions with length 1 and length 2,and the effects of different feature sets on the prediction of internal ions were investigated.