BiATNovo: A Self-Attention based Bidirectional Peptide Sequencing Method

Abstract

AbstractTraditional database-based peptide sequencing methods have shortcomings in discoverability and universality, while de novo sequencing is the essential way to analyze unknown proteins and discover new peptides and proteins. Most existing de novo sequencing algorithms have the problem of accumulated deviation and unbalanced output. At the same time, some algorithms could be more suitable for Data-Independent Acquisition Mass Spectrometry (DIA-MS) data. This paper designed a bidirectional peptide sequencing method to alleviate the problems of unbalanced output and deviation accumulation in the sequencing process. The self-attention mechanism was applied to de novo sequencing to increase the interaction within the peptide sequence and the interaction between the MS/MS spectra and the peptide sequence. On the DIA-MS dataset, the peptide prediction accuracy improved by an average of 15.6% compared with the state-of-the-art method. On the DDA-MS dataset, our method achieved the best performance on partial datasets, the amino acid accuracy improved by an average of 3%. At the same time, two new evaluation scores, Position-BLEU and Alignment score, were proposed to evaluate the misalignment between the predicted sequence and the reference sequence, and the partial absence of fragment ions.