Deephos: predicted spectral database search for TMT-labeled phosphopeptides and its false discovery rate estimation

Area of Science:

• Proteomics
• Cancer Research
• Bioinformatics

Background:

• Tandem mass spectrometry (MS/MS) with Tandem Mass Tag (TMT) labeling is crucial for quantifying post-translational modifications.
• Large-scale phosphopeptide quantification is vital in cancer proteogenomic studies.

Purpose of the Study:

• To develop a novel predicted Spectral DataBase (pSDB) search strategy, Deephos, to enhance the identification of TMT-labeled phosphopeptides.
• To improve both sensitivity and specificity in MS/MS spectrum identification for TMT-labeled phosphopeptides.

Main Methods:

• Compiled a pSDB of TMT-labeled phosphopeptides using deep learning-based fragment ion prediction from approximately 8000 human phosphoproteins.
• Developed an innovative method for generating decoy spectra to improve false discovery rate (FDR) estimation in pSDB searches.

Main Results:

• Deep learning accurately recognized fragmentation patterns affected by TMT labeling and phosphorylation.
• Demonstrated that existing decoy spectra generation methods can lead to inaccurate FDR estimation.
• Showcased the utility of Deephos in multi-stage analyses of glioblastoma, acute myeloid leukemia, and breast cancer phosphoproteomes.

Conclusions:

• Deephos offers a significant advancement in identifying TMT-labeled phosphopeptides.
• The proposed decoy spectra generation method ensures more accurate FDR estimation.
• Deephos enhances the analysis of cancer phosphoproteomes, contributing to proteogenomic research.