Identifying protein complexes in high-throughput protein interaction screens using an infinite latent feature model

Area of Science:

• Computational biology
• Systems biology
• Bioinformatics

Background:

• High-throughput protein-protein interaction (PPI) screens generate vast datasets.
• Identifying protein complexes and their constituents from PPI data is crucial for understanding cellular functions.
• Existing methods may require prior knowledge of the number of complexes, limiting discovery.

Purpose of the Study:

• To develop a novel Bayesian computational approach for identifying protein complexes from PPI screens.
• To enable proteins to be members of multiple complexes.
• To automatically infer the number of significant complexes from data without prior constraints.

Main Methods:

• A Bayesian framework integrating an infinite latent feature model for multi-complex membership.
• A graph diffusion kernel to assess the probability of proteins belonging to the same complex.
• Gibbs sampling for inferring a catalog of protein complexes from interaction data.

Main Results:

• The proposed method successfully infers a catalog of protein complexes.
• The model automatically determines the number of significant complexes.
• Validation using yeast RNA-processing complex data demonstrated biologically meaningful partitioning.

Conclusions:

• The Bayesian approach provides a powerful, flexible, and data-driven method for protein complex identification.
• This method enhances the understanding of protein complex organization and function.
• The approach is applicable to various high-throughput PPI datasets.