Abstract
In recent years, the amount of biomedical literatures grows rapidly and thus the need for automated relation extraction methods becomes critical. Among all types of relations, knowledge about protein–protein interactions, including information concerning various aspects of the structural and functional organization of cells, can shed light on molecular mechanisms of biological processes. Therefore, identifying the interactions between proteins mentioned in biomedical literatures is one of the frequently discussed topics of text mining in the life science field. In this paper we propose PIPE, an interaction pattern generation module used in BioCreative 2015 competition to capture frequent protein-protein interaction (PPI) patterns within text. We also present an interaction pattern tree kernel method that integrates the PPI patterns with convolution tree kernel to extract protein-protein interactions, and the interaction pattern tree is constructed through three operations including branching, pruning and ornamenting. The proposed tree structure incorporates syntactic, content, and semantic information in text. Methods were evaluated on LLL, IEPA, HPRD50, AIMed, and BioInfer corpora using cross-validation, cross-learning, and cross-corpus evaluation. Empirical evaluations demonstrate that our method is effective and outperforms several well-known PPI extraction methods. Moreover, we discuss further the features that may be useful for future research.