TIDD: tool-independent and data-dependent machine learning for peptide identification

Area of Science:

• Proteomics
• Computational Biology
• Mass Spectrometry

Background:

• Shotgun proteomics relies on database search engines and post-processing tools to identify peptides from MS/MS spectra.
• Popular tools like Percolator and PeptideProphet enhance peptide identification but require specific feature optimization for each search engine.
• This limitation hinders their application with novel or diverse search engine outputs.

Purpose of the Study:

• To develop a universal post-processing tool for peptide identification in shotgun proteomics.
• To overcome the limitations of existing methods that require engine-specific feature engineering.

Main Methods:

• Developed TIDD, a post-processing tool utilizing universal features to assess peptide-spectrum match quality.
• TIDD allows integration of additional search engine-specific features.
• Implemented a user-friendly graphical interface.

Main Results:

• TIDD demonstrated comparable or superior performance to Percolator in peptide identification.
• Achieved 10.23-38.95% increase in PSMs compared to target-decoy estimation for MSFragger.
• Successfully processed results from various search engines without prior feature optimization.

Conclusions:

• TIDD eliminates the need for specialized feature engineering for different database search tools.
• The tool is directly applicable to any database search results, including those from newly developed engines.
• Enhances confident peptide identifications across diverse proteomics search platforms.