JUPPI: A Multi-Level Feature Based Method for PPI Prediction and a Refined Strategy for Performance Assessment

Area of Science:

• Computational biology
• Bioinformatics
• Systems biology

Background:

• Protein-protein interactions (PPIs) are crucial for cellular processes.
• Existing computational PPI prediction methods face challenges with evaluation strategies and data selection.
• Accurate PPI prediction is essential for understanding biological functions and disease mechanisms.

Purpose of the Study:

• To develop a robust computational method for protein-protein interaction (PPI) prediction.
• To introduce a refined evaluation strategy addressing limitations in existing benchmarking approaches.
• To improve the prediction of interactions involving human proteins, including intrinsically disordered ones.

Main Methods:

• Proposed JUPPI (Just a Unified PPI Predictor) utilizing multi-level features: sequence, domain, and Gene Ontology (GO) information.
• Implemented a refined evaluation strategy involving high-quality negative data extraction and a pair-input cross-validation with three difficulty levels.
• Reduced component-level overlapping issues in test sets through the novel evaluation strategy.

Main Results:

• JUPPI demonstrated superior performance compared to state-of-the-art methods across six independent PPI datasets.
• The method achieved high accuracy in predicting PPIs at the human proteome level.
• JUPPI also excelled in predicting interactors for intrinsically disordered human proteins.

Conclusions:

• JUPPI offers a significant advancement in computational PPI prediction accuracy and reliability.
• The refined evaluation strategy provides a more rigorous benchmark for future PPI prediction methods.
• The JUPPI tool and datasets are publicly available, facilitating further research in the field.