Resumen de Semantic resources in pharmacovigilance: a corpus and an ontology for drug-drug interactions
María Herrero Zazo 
Nowadays, with the increasing use of several drugs for the treatment of one or more different diseases (polytherapy) in large populations, the risk for drugs combinations that have not been studied in pre-authorization clinical trials has increased. This provides a favourable setting for the occurrence of drug-drug interactions (DDIs), a common adverse drug reaction (ADR) representing an important risk to patients safety, and an increase in healthcare costs. Their early detection is, therefore, a main concern in the clinical setting. Although there are different databases supporting healthcare professionals in the detection of DDIs, the quality of these databases is very uneven, and the consistency of their content is limited. Furthermore, these databases do not scale well to the large and growing number of pharmacovigilance literature in recent years. In addition, large amounts of current and valuable information are hidden in published articles, scientific journals, books, and technical reports. Thus, the large number of DDI information sources has overwhelmed most healthcare professionals because it is not possible to remain up to date on everything published about DDIs. Computational methods can play a key role in the identification, explanation, and prediction of DDIs on a large scale, since they can be used to collect, analyze and manipulate large amounts of biological and pharmacological data. Natural language processing (NLP) techniques can be used to retrieve and extract DDI information from pharmacological texts, supporting researchers and healthcare professionals on the challenging task of searching DDI information among different and heterogeneous sources. However, these methods rely on the availability of specific resources providing the domain knowledge, such as databases, terminological vocabularies, corpora, ontologies, and so forth, which are necessary to address the Information Extraction (IE) tasks. In this thesis, we have developed two semantic resources for the DDI domain that make an important contribution to the research and development of IE systems for DDIs. We have reviewed and analyzed the existing corpora and ontologies relevant to this domain, based on their strengths and weaknesses, we have developed the DDI corpus and the ontology for drug-drug interactions (named DINTO). The DDI corpus has proven to fulfil the characteristics of a high-quality gold-standard, and has demonstrated its usefulness as a benchmark for the training and testing of different IE systems in the SemEval-2013 DDIExtraction shared task. Meanwhile, DINTO has been used and evaluated in two different applications. Firstly, it has been proven that the knowledge represented in the ontology can be used to infer DDIs and their different mechanisms. Secondly, we have provided a proof-of-concept of the contribution of DINTO to NLP, by providing the domain knowledge to be exploited by an IE pilot prototype. From these results, we believe that these two semantic resources will encourage further research into the application of computational methods to the early detection of DDIs. This work has been partially supported by the Regional Government of Madrid under the Research Network MA2VICMR [S2009/TIC-1542], by the Spanish Ministry of Education under the project MULTIMEDICA [TIN2010-20644-C03-01] and by the European Commission Seventh Framework Programme under TrendMiner project [FP7-ICT287863].
