UniReD

UniReD predicts protein-protein interactions by text-mining biomedical literature and integrating UniProtKB/Swiss-Prot annotations with PubMed-derived documents to extract known associations and propose novel ones.

Key Features:

• Text mining: Analyzes biomedical literature to identify associations between proteins.
• UniProtKB/Swiss-Prot integration: Leverages manually curated annotations from UniProtKB/Swiss-Prot to ground predictions in high-quality protein entries.
• PubMed document retrieval: Automatically identifies related documents from PubMed to expand literature-derived evidence.
• Clustering of proteins: Creates clusters of proteins putatively functionally related based on literature-derived associations.
• Candidate prioritization: Produces sorted lists of candidate proteins for potential interactions.
• Species specificity: Utilizes manually annotated protein entries specific to human and mouse.
• Experimental validation: Predicted interactions include six cases that were subsequently validated experimentally.
• Benchmarking and coverage: Achieved high coverage when compared against public databases of experimentally validated PPIs.
• Complement to experimental methods: Provides candidate lists that complement approaches such as mass spectrometry.

Scientific Applications:

• PPI discovery: Identification and suggestion of novel protein-protein interactions from the literature.
• Functional inference: Support for understanding cellular functions through analysis of protein association networks.
• Experimental prioritization: Prioritization of candidate proteins for experimental validation studies.
• Database augmentation: Expansion and cross-validation of PPI coverage relative to public experimentally validated databases.
• Complementary workflows: Integration with experimental workflows such as mass spectrometry to refine candidate selection.

Methodology:

Text mining of biomedical literature, integration of UniProtKB/Swiss-Prot manually curated annotations, automated identification of related PubMed documents, clustering of proteins based on literature-derived associations, and generation of sorted candidate protein lists.