MsTargetPeaker: A Quality-Aware Deep Reinforcement Learning Approach for Peak Identification in Targeted Proteomics.

Area of Science:

• Proteomics
• Analytical Chemistry
• Biotechnology

Background:

• Targeted mass spectrometry requires accurate peptide quantification through chromatographic peak integration.
• Automated peak identification is challenging for low-abundance peptides due to noise and interference.
• Current methods use separate supervised learning models for peak selection and quality control, limiting refinement.

Purpose of the Study:

• To develop a quality-aware search procedure for robust peak identification in targeted proteomics.
• To enhance the precision and accuracy of peptide quantification in mass spectrometry data.
• To provide interpretable diagnostic reports for efficient manual curation.

Main Methods:

• Implemented a reinforcement learning agent to guide Monte Carlo tree search for chromatogram exploration.
• Developed a custom reward function for dynamic peak quality assessment during search.
• Incorporated cross-sample consensus profiles to improve identification of ambiguous signals.

Main Results:

• MsTargetPeaker efficiently identifies target peaks while minimizing interference.
• Dynamic quality assessment enables accurate and robust peak boundary determination.
• The method enhances both peak quality and quantification precision for targeted proteomics.

Conclusions:

• MsTargetPeaker offers a practical advancement in automated peak identification for targeted proteomics.
• The quality-aware approach improves robustness and precision in peptide quantification.
• Generated diagnostic reports facilitate efficient manual review and data quality assessment.