EBprot: Statistical analysis of labeling-based quantitative proteomics data.

Area of Science:

• Proteomics and Bioinformatics
• Quantitative Mass Spectrometry
• Statistical Modeling in Biology

Background:

• Labeling-based proteomics is crucial for identifying differentially expressed proteins (DEPs).
• Current methods often summarize peptide-level data to protein-level ratios, neglecting peptide reproducibility.
• Shotgun proteomics exhibits variable peptide counts per protein, impacting differential expression (DE) analysis reliability.

Purpose of the Study:

• To develop a novel probabilistic framework, EBprot, for direct peptide-protein hierarchy modeling.
• To enhance differential expression analysis by incorporating peptide-level reproducibility.
• To provide a robust alternative to existing statistical methods for quantitative proteomics data.

Main Methods:

• Proposed EBprot, a probabilistic framework modeling the peptide-protein hierarchy.
• Utilized simulation studies to compare EBprot's peptide-level analysis with protein-level ratio methods.
• Evaluated EBprot on a spike-in dataset and applied it to lung cancer and phosphoproteome datasets.

Main Results:

• EBprot's peptide-level analysis demonstrated superior receiver-operating characteristic performance and more accurate false discovery rate estimation in simulations.
• EBprot showed enhanced classification performance on a spike-in dataset.
• EBprot proved robust across diverse experimental designs, including subtype analysis and time-course phosphoproteomics.

Conclusions:

• EBprot offers a robust peptide-level analysis for labeling-based quantitative proteomics.
• The framework effectively incorporates peptide reproducibility, improving differential expression detection.
• EBprot is a valuable alternative to traditional protein-level ratio-based statistical methods.