Improving protein interaction prediction in GenPPi: a novel interaction sampling approach preserving network topology

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Protein-protein interactions (PPIs) are vital for cellular processes and drug discovery.
• Existing computational methods like GenPPi were limited by high sequence similarity requirements.
• GenPPi 1.5 is developed to overcome these limitations and enhance predictive capabilities.

Purpose of the Study:

• To improve ab initio prediction of PPI networks, especially for genomes with low sequence similarity.
• To integrate advanced machine learning and sampling algorithms for more accurate and efficient PPI prediction.
• To provide a scalable and customizable tool for analyzing protein interaction networks.

Main Methods:

• Incorporation of a Random Forest (RF) algorithm trained on 60 biophysical features to classify protein similarity.
• Development and integration of the Reduced Interaction Sampling (RIS) algorithm to manage computational complexity.
• Validation through extensive simulations and application to bacterial genomes like Buchnera aphidicola.

Main Results:

• GenPPi 1.5 demonstrates improved classification of protein similarity, even at lower sequence identities.
• The RF model significantly expanded predictive power, showing up to 62% overlap with known interactions (e.g., STRING database).
• RIS algorithm efficiently handles complex genomes while maintaining robust identification of critical network nodes.

Conclusions:

• GenPPi 1.5 offers a significant advancement in alignment-free PPI prediction by combining RF and RIS.
• The enhanced tool overcomes previous limitations, providing a robust and scalable solution for diverse genomic contexts.
• GenPPi 1.5 is freely available, user-friendly, and allows for custom model training.