Phosphoproteome Prediction

Phosphoproteome Prediction predicts phosphorylation levels of proteins across cancer patients from proteomic, transcriptomic, and genomic datasets as an in silico alternative to mass spectrometry-based phosphoproteomics.

Key Features:

• Input data integration: Forecasts phosphoproteomic profiles by leveraging proteomic, transcriptomic, and genomic datasets.
• Baseline Correlations: Utilizes baseline correlations between protein abundances and corresponding phosphoprotein levels to inform predictions.
• Universal Protein-Protein Interactions: Incorporates universal protein-protein interactions to account for network-level influences on phosphorylation.
• Shareable Regulatory Information Across Cancer Tissues: Harnesses regulatory information shared across different cancer tissues to improve generalization.
• Associations Among Multi-Phosphorylation Sites: Considers associations among multiple phosphorylation sites on the same protein to capture site interdependencies.
• Validation on held-out cohorts: Evaluated on a held-out dataset of 108 breast cancer samples and 62 ovarian cancer samples, achieving top predictive accuracy among compared methods.
• Origin: Developed as part of a winning entry in the 2017 NCI-CPTAC DREAM Proteogenomics Challenge.

Scientific Applications:

• In silico phosphoproteomics: Provides predicted phosphoproteomic profiles as an alternative to experimental mass spectrometry-based phosphoproteomics.
• Regulatory mechanism investigation: Enables study of protein phosphorylation regulatory mechanisms across cancer tissues.
• Oncogenesis research: Facilitates analysis of how aberrant phosphorylation events contribute to cancer development and progression.

Methodology:

Integrates proteomic, transcriptomic, and genomic datasets and combines baseline protein–phosphoprotein correlations, universal protein-protein interactions, shareable regulatory information across cancer tissues, and associations among multi-phosphorylation sites.