Sieve-based relation extraction of gene regulatory networks from biological literature

Žitnik et al. BMC Bioinformatics 2015, 16(Suppl 16):S1 http://www.biomedcentral.com/1471-2105/16/S16/S1 RESEARCH Open Access Sieve-based relation extraction of gene regulatory networks from biological literature Slavko Žitnik1,3*, Marinka Žitnik1, Blaž Zupan1,2, Marko Bajec1 From BioNLP Shared Task 2013 Sofia, Bulgaria. 9 August 2013 Abstract Background: Relation extraction is an essential procedure in literature mining. It focuses on extracting semantic relations between parts of text, called mentions. Biomedical literature includes an enormous amount of textual descriptions of biological entities, their interactions and results of related experiments. To extract them in an explicit, computer readable format, these relations were at first extracted manually from databases. Manual curation was later replaced with automatic or semi-automatic tools with natural language processing capabilities. The current challenge is the development of information extraction procedures that can directly infer more complex relational structures, such as gene regulatory networks. Results: We develop a computational approach for extraction of gene regulatory networks from textual data. Our method is designed as a sieve-based system and uses linear-chain conditional random fields and rules for relation extraction. With this method we successfully extracted the sporulation gene regulation network in the bacterium Bacillus subtilis for the information extraction challenge at the BioNLP 2013 conference. To enable extraction of distant relations using first-order models, we transform the data into skip-mention sequences. We infer multiple models, each of which is able to extract different relationship types. Following the shared task, we conducted additional analysis using different system settings that resulted in reducing the reconstruction error of bacterial sporulation network from 0.73 to 0.68, measured as the slot error rate between the predicted and the reference network. We observe that all relation extraction sieves contribute to the predictive performance of the proposed approach. Also, features constructed by considering mention words and their prefixes and suffixes are the most important features for higher accuracy of extraction. Analysis of distances between different mention types in the text shows that our choice of transforming data into skip-mention sequences is appropriate for detecting relations between distant mentions. Conclusions: Linear-chain conditional random fields, along with appropriate data transformations, can be efficiently used to extract relations. The sieve-based architecture simplifies the system as new sieves can be easily added or removed and each sieve can utilize the results of previous ones. Furthermore, sieves with conditional random fields can be trained on arbitrary text data and hence are applicable to broad range of relation extraction tasks and data domains. Background We are witnessing an unprecedented increase in the number of biomedical abstracts, experimental results and phenotype and gene descriptions being deposited to publicly available databases, such as NCBI’s PubMed. * Correspondence: 1 Faculty of Computer and Information Science, University of Ljubljana, Večna pot 113, SI-1000 Ljubljana, Slovenia Full list of author information is available at the end of the article Collectively, this content represents potential new discoveries that could be inferred with appropriately designed natural language processing approaches. Identification of topics that appear in biomedical research literature was among first computational approaches to predict associations between diseases and genes and has become indispensable to both researchers in the biomedical field and curators [1-4]. Information from publication repositories is often mined together with other data © 2015 Žitnik et al. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http:// creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/ zero/1.0/) applies to the data made available in this article, unless otherwise stated. Žitnik et al. BMC Bioinformatics 2015, 16(Suppl 16):S1 http://www.biomedcentral.com/1471-2105/16/S16/S1 sources. Databases that store relations from integrative mining are for example the OMIM database on human genes and genetic phenotypes [5], the GeneRIF function annotation database [6], the Gene Ontology [7] and clinical drug information from the DailyMed database [8]. Biomedical mining of literature is a compelling way to identify possible candidate genes through integration of existing data. A dedicated set of computational techniques is required to infer structured relations from plain textual information stored in large literature databases [9]. Relation extraction tools [10] can identify semantic relations between entities found in text. Early relationship extraction systems relied mostly on manually defined rules to extract a limited number of relationship types [11]. Later, machine learning-based methods were introduced to address the extraction task by inferring prediction models from sets of labeled relationship types [12-14]. When no labeled data were available, unsupervised systems were developed to extract relationship descriptors based on the language syntax [10]. Current state-of-the-art systems combine both machine learning and rule-based approaches to extract relevant information from narrative summaries and represent it in a structured form [15,16]. This paper aims at the extraction of gene regulatory networks of Bacillus subtilis. The reconstruction and elucidation of gene regulation networks is an important task that can change our understanding of the processes and molecular interactions within the cell [17-19]. We have developed a novel sieve-based computational methodology that builds upon conditional random fields [20] and specialized rules to extract gene relations from unstructured text. Extracted relations are assembled into a multi-relational gene network that is informative of the type of regulation between pairs of genes and the directionality of their action. The proposed approach can consider biological literature on gene interactions from multiple data sources. The main novelty of our work here is the construction of a sequential analysis pipeline for extracting gene relations of various types from literature data (Figure 1). We demonstrate the effectiveness and applicability of our recently proposed coreference resolution system [21]. Our system uses linear-chain conditional random fields in an innovative way and can detect distant coreferent mentions in text using a novel transformation of data into skip-mention sequences. We evaluate the proposed methodology by measuring th (...truncated)