PepGo: a deep learning and tree search-based model for de novo peptide sequencing

Abstract

Identifying peptide sequences from tandem mass spectra is a fundamental problem in proteomics. Unlike search-based methods that rely on matching spectra to databases, de novo peptide sequencing determines peptides directly from mass spectra without any prior information. However, the design of models and algorithms for de novo peptide sequencing remains a challenge. Many de novo approaches leverage deep learning but primarily focus on the architecture of neural networks, paying less attention to search algorithms. We introduce PepGo, a de novo peptide sequencing model that integrates Transformer neural networks with Monte Carlo Tree Search (MCTS). PepGo predicts peptide sequences directly from mass spectra without databases, even without prior training. We show that PepGo surpasses existing methods, achieving state-of-the-art performance. To our knowledge, this is the first approach to combine deep learning with MCTS for de novo peptide sequencing, offering a powerful and adaptable solution for peptide identification in proteomics research.