Deep Learning Methods for De Novo Peptide Sequencing.

Area of Science:

• Proteomics and Bioinformatics
• Computational Biology
• Mass Spectrometry Data Analysis

Background:

• Protein tandem mass spectrometry is typically interpreted using reference genome databases.
• Challenges arise when these protein databases are unavailable or incomplete, necessitating de novo sequencing.
• Deep learning has become dominant in de novo sequencing since the DeepNovo algorithm in 2017.

Purpose of the Study:

• To describe deep learning methods used in de novo sequencing.
• To outline performance evaluation procedures for these methods.
• To discuss current challenges in de novo sequencing methods development and evaluation.

Main Methods:

• Review of deep learning algorithms for translating mass spectra to peptide sequences.
• Examination of established protocols for assessing the accuracy and efficiency of de novo sequencing tools.
• Analysis of common benchmarks and datasets used in the field.

Main Results:

• Deep learning models, trained on extensive labeled mass spectrometry data, show significant promise in de novo sequencing.
• Performance evaluation highlights the strengths and limitations of current deep learning approaches.
• Identified key areas for improvement in both algorithmic development and standardized evaluation metrics.

Conclusions:

• Deep learning represents a powerful paradigm for de novo sequencing in proteomics.
• Standardized evaluation protocols are crucial for advancing the field.
• Continued research is needed to address challenges in methods development and ensure robust performance.