Application of information extraction techniques to pharmacological domainextracting drug-drug interactions

Zusammenfassung

A drug-drug interaction occurs when one drug influences the level or activity of another drug. The detection of drug interactions is an important research area in patient safety since these interactions can become very dangerous and increase health care costs. Although there are different databases supporting health care professionals in the detection of drug interactions, this kind of resource is rarely complete. Drug interactions are frequently reported in journals of clinical pharmacology, making medical literature the most effective source for the detection of drug interactions. However, the increasing volume of the literature overwhelms health care professionals trying to keep an up-to-date collection of all reported drug-drug interactions. The development of automatic methods for collecting, maintaining and interpreting this information is crucial to achieving a real improvement in their early detection. Information Extraction techniques can provide an interesting way to reduce the time spent by health care professionals on reviewing the literature. Nevertheless, only a few approaches have tackled the extraction of drug-drug interactions. In this thesis, we have conducted a detailed study about various information extraction techniques applied to biomedical domain. Based on this study, we have proposed two different approximations for the extraction of drug-drug interactions from texts. The first approximation proposes a hybrid approach, which combines shallow parsing and pattern matching to extract relations between drugs from biomedical texts. The second approximation is based on a supervised machine learning approach, in particular, kernel methods. In addition, we have created and annotated the first corpus, DrugDDI, annotated with drug-drug interactions, which allow us to evaluate and compare both approximations. We think the DrugDDI corpus is an important contribution because it could encourage other research groups to investigate in this problem. To the best of our knowledge, the DrugDDI corpus is the only available corpus annotated for drug-drug interactions and this thesis is the first work which addresses the problem of extracting drug-drug interactions from biomedical texts. We have also defined three auxiliary processes to provide crucial information, which will be used by the aforementioned approximations. These auxiliary tasks are as follows: (1) a process for text analysis based on the UMLS MetaMap Transfer tool (MMTx) to provide shallow syntactic and semantic information from texts, (2) a process for drug name recognition and classification, and (3) a process for drug anaphora resolution. Finally, we have developed a pipeline prototype which integrates the different auxiliary processes. The pipeline architecture allows us to easily integrate these modules with each of the approaches proposed in this thesis: pattern-matching or kernels. Several experiments were performed on the DrugDDI corpus. They show while the first approximation based on pattern matching achieves low performance, the approach based on kernel-methods achieves a performance comparable to those obtained by approaches which carry out a similar task as the extraction of protein-protein interactions.