A learned score function improves the power of mass spectrometry database search

Area of Science:

• Proteomics
• Computational Biology
• Mass Spectrometry

Background:

• Peptide assignment is crucial for analyzing protein tandem mass spectrometry data.
• Current methods include database search and de novo sequencing.
• De novo sequencing often uses machine learning, while database search uses hand-designed scoring functions.

Purpose of the Study:

• To test the hypothesis that machine learning models for de novo sequencing can be repurposed as scoring functions for database search.
• To develop a novel scoring function for peptide identification in mass spectrometry.

Main Methods:

• Re-engineered the Casanovo tool, a state-of-the-art de novo sequencing method, to score peptide-spectrum pairs.
• Evaluated the performance of the new scoring function, Casanovo-DB, on a benchmark dataset.
• Assessed the impact of the Percolator post-processor on Casanovo-DB performance.

Main Results:

• The Casanovo-DB scoring function demonstrated statistical power in detecting peptides across different species.
• Re-scoring with Percolator further improved the performance of Casanovo-DB.
• The machine learning-derived score function showed potential to outperform traditional methods.

Conclusions:

• Machine learning models from de novo sequencing can be effectively repurposed for peptide scoring in database search.
• Casanovo-DB offers a promising alternative to hand-designed scoring functions in proteomics.
• The integration with Percolator enhances the utility of Casanovo-DB for peptide identification.