Repurposing of drug label information to create actionable intelligence with BioWisdom's Metawise.
Paul Bradley and Jane Reed, BioWisdom, UK
Drug toxicity is now the major cause of drug failure in clinical development. As a result, there is currently a focus on developing early screening methods to predict potential toxicities so that safer compounds can be selected, or toxic liabilities can be engineered out of promising candidates. The complexity of drug toxicity is reflected in the variety of approaches being developed, from conventional in vitro screens to computational QSAR methods and other in silico models. The industry is also looking at ways to reduce costs by improving R&D interoperability and streamlining the pre-competitive elements of the drug discovery process.
Improved toxicological prediction demands the best view of current and historic data. There are many barriers along the path to better data repurposing, pre-competitive sharing, and harmonisation – all key goals in the current climate of translational healthcare. Achieving this goal is made more challenging by the variety of ways in which different communities describe important concepts within the healthcare domain, such as drugs, targets, tissues, and diseases. We will introduce Metawise, a novel approach to terminology management that provides flexible and controllable terminology for search and identification of life science concepts from text sources. Metawise identifies key terms in text using a unique approach based on term structure and semantics, which enables the system to recognise the varied language that domain specialists use to refer to important concepts and provides a mechanism for their harmonisation to any preferred standard. This allows common standards, such as the Standard for Exchange of Non-Clinical Data (SEND), to be more easily adopted and improves the flow of data between organisations.
We will present a use case in which Metawise was used to extract intelligence from the FDA's DailyMed drug package inserts. The work was undertaken on behalf of a US Government Agency to provide decision support to physicians. As of May 2011, DailyMed featured labels for around 24,000 drug products submitted to the FDA. These documents are composed of tagged sections such as indications and usage, adverse reactions, contraindications and boxed warnings. However, the text within each section, while rich in information, is provided as a block of unstructured text with little tagging or mark-up of important concepts such as drug and disease names. Furthermore, labels are compiled without reference to a universal terminology standard, leading to the widespread use of related terms such as “hemorrhage/bleeding” and "sleeplessness/insomnia". Metawise was able to identify and tag references to drug safety and translate clinical terms to their SNOMED CT equivalent, thereby providing a consistent standard across labels. The concepts and relationships that were recognised by Metawise were combined in the form of subject-verb-object triplets to create assertional metadata. This type of metadata summarises the key facts contained within text and creates a semantically normalised, actionable information layer over documents.
