Deep representation features from DreamDIA^XMBD improve the analysis of data-independent acquisition proteomics

Area of Science:

• Proteomics
• Bioinformatics
• Computational Biology

Background:

• Data-independent acquisition (DIA) mass spectrometry is a powerful technique for large-scale proteomic analysis.
• Accurate peptide identification and quantification are crucial for biological discovery using DIA data.
• Existing DIA analysis tools have limitations in maximizing peptide coverage and maintaining specificity.

Purpose of the Study:

• To develop a novel software suite, DreamDIA, for enhanced data-independent acquisition mass spectrometry analysis.
• To improve peptide identification and quantification performance in DIA datasets.
• To provide a publicly available tool for high-coverage and high-accuracy DIA data analysis.

Main Methods:

• Developed DreamDIA, a software suite utilizing a deep representation model for DIA data analysis.
• Employed a data-driven strategy to capture comprehensive elution pattern information from peptides.
• Extracted additional features from hundreds of theoretical elution profiles using a deep representation network.
• Utilized nonlinear discriminative models within a positive-unlabeled learning framework with decoy peptides for enhanced specificity.

Main Results:

• DreamDIA demonstrated considerable improvements in both peptide identification and quantification performance.
• Achieved superior results compared to state-of-the-art methods like OpenSWATH, Skyline, and DIA-NN.
• Enabled higher coverage of target peptides without sacrificing analytical specificity.

Conclusions:

• DreamDIA offers a significant advancement in DIA data analysis software.
• The deep representation model and positive-unlabeled learning approach enhance performance.
• DreamDIA is a valuable, publicly accessible resource for the proteomics community.