Improving X!Tandem on peptide identification from mass spectrometry by self-boosted Percolator

Area of Science:

• Proteomics
• Computational Biology
• Bioinformatics

Background:

• Accurate protein identification is crucial for mass spectrometry (MS)-based proteomics.
• Peptide-spectrum matches (PSMs) from database search algorithms require validation for reliable downstream analysis.
• Existing popular postprocessing algorithms like Percolator and PeptideProphet are primarily designed for commercial search engines (SEQUEST, MASCOT).

Purpose of the Study:

• To extend and optimize PSM postprocessing algorithms for open-source database search algorithms, specifically X!Tandem.
• To introduce a novel Self-boosted Percolator method for enhancing PSM validation in X!Tandem results.
• To improve the performance of semi-supervised learning (SSL) algorithms by addressing their sensitivity to initial PSM ranking.

Main Methods:

• Proposed a Self-boosted Percolator approach for postprocessing X!Tandem search results.
• Implemented Percolator in a cascade learning manner to progressively improve performance.
• Evaluated the method's ability to increase PSM identifications while maintaining a controlled false discovery rate (FDR).

Main Results:

• Demonstrated that the SSL algorithm in Percolator is highly dependent on the initial ranking of PSMs.
• Showcased that a cascade learning approach (Self-boosted Percolator) can overcome suboptimal initial rankings.
• Achieved improved PSM identification rates without compromising the false discovery rate (FDR) when applied to X!Tandem.

Conclusions:

• Self-boosted Percolator offers a significant improvement for PSM validation with open-source search algorithms like X!Tandem.
• The cascade learning strategy enhances the robustness and performance of SSL-based postprocessing.
• This optimized approach enables more comprehensive protein identification in MS-based proteomics studies.