Protein-Protein Interactions Prediction via Multimodal Deep Polynomial Network and Regularized Extreme Learning...

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Protein-protein interactions (PPIs) are crucial for numerous biological processes.
• Accurate prediction of PPIs is essential for various applications in biology and medicine.
• Existing computational methods for PPI prediction require improvement.

Purpose of the Study:

• To develop a novel computational method for predicting protein-protein interactions (PPIs).
• To effectively integrate diverse protein features for enhanced prediction accuracy.
• To introduce a multimodal deep polynomial network (MDPN) for PPI prediction.

Main Methods:

• A multimodal deep polynomial network (MDPN) was proposed, featuring a two-stage deep polynomial network (DPN).
• The first stage encodes multiple protein features into high-level representations.
• The second stage fuses and learns from these features using a regularized extreme learning machine for PPI prediction.

Main Results:

• The MDPN achieved high average accuracies: 97.87% for H. pylori, 99.90% for Human, and 98.11% for Yeast.
• The method demonstrated robust performance on various datasets and PPI networks.
• Superior prediction results were obtained compared to existing methods.

Conclusions:

• The proposed MDPN effectively integrates protein mutation rates and physicochemical properties for accurate PPI prediction.
• This novel computational approach significantly enhances the performance of PPI prediction.
• MDPN offers a promising tool for advancing research in protein interaction networks.