Refining literature curated protein interactions using expert opinions

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• High-quality physical protein-protein interaction (PPI) datasets are crucial for interactome analysis and predictive modeling.
• Literature-curated PPI databases often lack clear labeling of physical interactions, impacting dataset quality.
• Accurate mapping of human immunodeficiency virus (HIV-1) interactions with host proteins is essential for understanding viral mechanisms.

Purpose of the Study:

• To develop a robust framework for generating high-quality gold standard datasets of physical PPIs.
• To specifically curate a reliable dataset of physical interactions between HIV-1 and human proteins.
• To provide a method applicable to refining other existing biological databases.

Main Methods:

• A crowd-sourcing approach was employed, collecting expert opinions on protein-protein associations.
• An expectation-maximization algorithm was used to estimate the quality of expert labeling.
• Probabilistic inference was applied to determine the likelihood of a reported PPI being a direct physical interaction.

Main Results:

• A high-quality physical interaction network between HIV-1 and human proteins was successfully obtained.
• The effectiveness of the crowd-sourcing and probabilistic framework was validated using synthetic data.
• The curated dataset is publicly available for research use.

Conclusions:

• The developed framework effectively addresses challenges in curating high-quality physical PPI data.
• This approach can significantly enhance the reliability of interactome databases.
• The methodology offers a valuable tool for refining biological interaction datasets, particularly for viral-host interactions.