PPI finder: a mining tool for human protein-protein interactions

Area of Science:

• Biomedical Informatics
• Computational Biology
• Bioinformatics

Background:

• The rapid growth of biomedical literature necessitates automated text mining for information extraction.
• Protein-protein interactions (PPIs) are commonly mined from PubMed abstracts using co-occurrence or rule-based methods.
• Hybrid approaches combining methods may improve PPI prediction, but require robust validation against databases and ontologies.

Purpose of the Study:

• To develop and present PPI Finder, a web-based tool for mining human PPIs from PubMed abstracts.
• To evaluate the performance of PPI prediction using co-occurrence and interaction words, validated by PPI databases and Gene Ontology (GO) terms.

Main Methods:

• Developed PPI Finder, a web tool integrating co-occurrence analysis, interaction word identification, and validation against human PPI databases (HPRD, BioGRID, BIND) and GO terms.
• Assessed the overlap between co-occurring pairs in abstracts and known PPI databases.
• Examined the co-occurrence and GO term similarity for known PPIs.

Main Results:

• Only 28% of PubMed co-occurring pairs were found in common human PPI databases.
• Conversely, 69% of known PPIs from HPRD exhibited co-occurrence in the literature.
• A significant 65% of known PPIs shared common terms within the Gene Ontology database.

Conclusions:

• PPI Finder offers a valuable resource for biologists to discover potential novel protein-protein interactions.
• The tool facilitates the identification of PPIs by leveraging literature-based evidence and curated databases.
• PPI Finder is freely accessible online for research purposes.