Integrating semantic information into multiple kernels for protein-protein interaction extraction from biomedical...

Area of Science:

• Biomedical Informatics
• Computational Biology
• Natural Language Processing

Background:

• Protein-Protein Interaction (PPI) extraction is crucial for understanding biological processes.
• Current machine learning methods for PPI extraction show promise but require improvement.
• Existing methods often overlook valuable semantic information.

Purpose of the Study:

• To develop an advanced multiple-kernel learning approach for more accurate PPI extraction.
• To integrate diverse features, including syntactic and semantic information, into a unified model.
• To leverage semantic resources like WordNet and MeSH to enhance PPI extraction performance.

Main Methods:

• Proposed a multiple-kernel learning framework combining feature-based, tree, and semantic kernels.
• Extended the shortest path-enclosed tree kernel (SPT) with a dynamic strategy for richer syntactic information.
• Developed a semantic kernel to compute protein-protein pair and context similarity using WordNet and MeSH.

Main Results:

• Achieved an F-score of 69.40% and an Area Under the Curve (AUC) of 92.00% using Support Vector Machine (SVM).
• Demonstrated superior performance compared to most state-of-the-art systems.
• Highlighted the significant contribution of integrating semantic information.

Conclusions:

• The proposed multiple-kernel learning approach effectively integrates diverse features for improved PPI extraction.
• Incorporating semantic resources significantly enhances the accuracy of biomedical information extraction.
• This method offers a robust solution for advancing PPI extraction in bioinformatics.