Integrating PPI datasets with the PPI data from biomedical literature for protein complex detection

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Protein complexes are crucial for cellular functions.
• High-throughput experiments generate vast protein-protein interaction (PPI) data.
• Biomedical literature is an underutilized source of PPI information.

Purpose of the Study:

• To develop an approach for integrating PPI data from biomedical literature into existing PPI datasets.
• To enhance protein complex detection by leveraging both experimental and literature-derived interaction data.

Main Methods:

• Utilized a natural language processing system (PPIExtractor) to extract PPIs from biomedical literature.
• Integrated extracted literature PPIs with existing PPI datasets.
• Applied state-of-the-art complex detection algorithms (ClusterONE, CMC, COACH, RRW) to the integrated networks.

Main Results:

• Demonstrated significant improvements in protein complex detection accuracy (e.g., 3.976-5.416 percentage units in Maximum Matching Ratio).
• Validated the effectiveness of the integration approach across multiple yeast PPI datasets and gold standards (MIPS, SGD).
• Confirmed the approach's compatibility with various protein complex detection algorithms.

Conclusions:

• Biomedical literature is a valuable and accessible resource for PPI data.
• Integrating literature-derived PPIs into networks substantially improves protein complex detection performance.
• This strategy offers a robust method for enhancing our understanding of cellular organization and function.