De Novo Peptide Sequencing Based On Deep Learning And Ion Information Of Tandem Mass Spectrometry Spectra

In recent years,with the development of molecular biology and mass spectrometry,proteomics has drawn more and more attention and has been extensively applied in many fields.As one of the key technologies in proteomics,de novo peptide sequencing has irreplaceable advantages compared with other protein identification methods,which can extract peptide sequences directly from tandem mass spectrometry(MS/MS)spectra without any protein databases.Since the accuracy and efficiency of de novo peptide sequencing can be affected by the quality of the MS/MS data,the DeepNovo method using deep learning for de novo peptide sequencing is introduced,which outperforms the other state-of-the-art de novo sequencing methods in both amino acid level and peptide level.And the results of the DeepNovo method indicate that deep learning plays an important role in mass spectrometry and protein identification.However,there are still some shortcomings in the DeepNovo method.On the one hand,only b_ions and y_ions in the MS/MS spectra are considered in DeepNovo method,which may affect the performance of de novo peptide sequencing.On the other hand,since the DeepNovo method uses a fixed number of training iterations,the training model is not adaptive,and it is unfavorable to improve the accuracy of de novo peptide sequencing.For superior performance and better generalization ability,two improvements are mainly introduced in this paper for the DeepNovo method based on deep learning and ion information in the MS/MS spectra: a_ions in the MS/MS spectra are added to consider more fragment ions in the MS/MS spectra as well as their relationships for higher accuracy of de novo peptide sequencing,and the validation set is introduced to automatically determine the number of training iterations in deep learning,which is used to ensure that the training model is self-adaptive.Experimental results show that compared to the DeepNovo method,adding a_ions in the MS/MS spectra and introducing the validation set can both improve the performance of de novo sequencing effectively.In addition,the DeepNovo A+ method proposed in this paper is the best-performing method,because it combines both a_ions in the MS/MS spectra and the validation set during de novo peptide sequencing.And,as the peptide gets longer,the advantage of the DeepNovo A+ method is more obvious.