Ensemble learning prediction of protein-protein interactions using proteins functional annotations.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in biology

Background:

• Protein-protein interactions (PPIs) are crucial for biological processes.
• Existing PPI prediction methods rely on scattered data from multiple databases.
• There is a need for consolidated, high-throughput datasets for accurate PPI prediction.

Purpose of the Study:

• To create and evaluate novel high-throughput datasets of interacting proteins.
• To develop and assess an Ensemble Learning method for predicting PPIs.
• To provide processed PPI datasets and software for the research community.

Main Methods:

• Extracted PPI data from DIP, MINT, BioGRID, and IntAct databases.
• Generated features using Gene Ontology and DOMINE data for machine learning.
• Implemented an Ensemble Learning method using Support Vector Machine, Random Forest, Decision Tree, and Naïve Bayes with majority voting.

Main Results:

• The Ensemble Learning method achieved over 80% sensitivity and 90% accuracy in cross-validation.
• Performance on a larger, realistic dataset maintained over 70% sensitivity.
• The developed datasets are suitable for PPI prediction tasks.

Conclusions:

• The created datasets and Ensemble Learning method are effective for protein-protein interaction prediction.
• Ensemble Learning demonstrates strong suitability for computational prediction of PPIs.
• The processed datasets and software are publicly available to facilitate research.