Integrating Semantic Information into Multiple Kernels for Protein-Protein Interaction Extraction from Biomedical Literatures

• Published in: PloS one Vol. 9; no. 3; p. e91898
• Main Authors: Li, Lishuang, Zhang, Panpan, Zheng, Tianfu, Zhang, Hongying, Jiang, Zhenchao, Huang, Degen
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 12.03.2014; Public Library of Science (PLoS)
• Subjects: Algorithms; Artificial intelligence; Automation; Bioinformatics; Biology; Biomedical Research; Computer Science; Feature extraction; Information retrieval; Kernels; Keywords; Learning algorithms; Linguistics; Machine learning; Medical research; Medical Subject Headings; Natural language; Natural Language Processing; Ontology; Protein interaction; Protein Interaction Mapping; Protein-protein interactions; Proteins; Semantics; Shortest-path problems; Similarity; Studies; Support vector machines; Teaching methods; Trees
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0091898

Protein-Protein Interaction (PPI) extraction is an important task in the biomedical information extraction. Presently, many machine learning methods for PPI extraction have achieved promising results. However, the performance is still not satisfactory. One reason is that the semantic resources were basically ignored. In this paper, we propose a multiple-kernel learning-based approach to extract PPIs, combining the feature-based kernel, tree kernel and semantic kernel. Particularly, we extend the shortest path-enclosed tree kernel (SPT) by a dynamic extended strategy to retrieve the richer syntactic information. Our semantic kernel calculates the protein-protein pair similarity and the context similarity based on two semantic resources: WordNet and Medical Subject Heading (MeSH). We evaluate our method with Support Vector Machine (SVM) and achieve an F-score of 69.40% and an AUC of 92.00%, which show that our method outperforms most of the state-of-the-art systems by integrating semantic information.