PRER

PRER generates patient representations by computing pairwise relative protein expression features within protein-protein interaction (PPI) networks to improve prognostic models such as survival prediction.

Key Features:

• Integration with PPI Networks: Combines individual protein expression levels with protein-protein interaction (PPI) network context.
• Pairwise Relative Expression Analysis: Computes patient feature vectors from pairwise comparisons between a source protein's expression and the expressions of its neighboring proteins.
• Biased Random-Walk Strategy: Uses a biased random-walk on the PPI network to determine the neighborhood of each protein for feature computation.
• Survival Prediction Performance: Evaluated across 10 cancer types using random forest survival models, demonstrating statistically significant improvement in 9 of 10 cancers compared to representations based solely on individual protein expression.
• Identification of Novel Biomarkers: Reveals proteins that are predictive in models trained with PRER features but may be missed by conventional single-protein analyses.

Scientific Applications:

• Survival prediction in cancer: Improves survival prediction models across multiple cancer types by leveraging protein expression within PPI networks.
• Biomarker discovery: Identifies candidate prognostic proteins for personalized medicine and targeted therapy strategies.
• Extension to other diseases: Applicable to prognostic and molecular profiling analyses in other complex diseases that rely on molecular expression profiles.

Methodology:

Integrating protein expression data with PPI networks; computing pairwise relative-expression feature vectors using a biased random-walk to define protein neighborhoods; evaluating performance with survival prediction models such as random forests.