MULTIPROSPECTOR: an algorithm for the prediction of protein-protein interactions by multimeric threading.

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Systems Biology

Background:

• Identifying protein-protein interactions (PPIs) is crucial for understanding physiological functions in the postgenomic era.
• The growing number of solved protein complex structures enables the prediction of quaternary structures using protein threading methods.

Purpose of the Study:

• To develop and validate a multimeric threading approach for predicting protein-protein interactions and quaternary structures on a genomic scale.
• To enhance the accuracy of PPI prediction compared to existing sequence-based methods.

Main Methods:

• A two-phase threading approach: first, traditional single-chain threading using PROSPECTOR, then re-threading on known complexes including interfacial energy.
• Construction of a multimeric protein structure database and derivation of interfacial pairwise potentials.
• Development of empirical indicators based on threading Z-score and interfacial energy to identify true multimers.

Main Results:

• The algorithm correctly identified 36/40 homodimers, 15/15 heterodimers, and 65/69 monomers in a benchmark test set.
• In yeast protein interaction analysis, the multimeric threading algorithm predicted 144 interacting proteins, significantly outperforming PSI-BLAST (56 or 26 predictions).
• The method predicted 2865 PPIs in yeast, with 1138 validated against the Database of Interacting Proteins (DIP).

Conclusions:

• The developed multimeric threading approach is a promising tool for assisting in the assignment of protein-protein interactions on a genomic scale.
• This method offers improved accuracy and coverage for predicting PPIs and quaternary structures compared to sequence homology methods.