Computational LOINC to Support Biomedical Research at Scale

点击翻译按钮获取中文摘要

A core requirement for modern data science is the annotation of data and datasets to support linkage, indirect reference, and reasoning across domain specific knowledgebases. Clinical laboratory data must be annotated with standard reference concepts to seamlessly play its part in data-science analytics. For over 25 years, the Logical Observation Identifiers Names and Codes (LOINC®) terminology standard from the Regenstrief Institute has played the role of trusted identifiers for many clinical observations. LOINC codes are logically composed from constituent Parts to describe unique concepts with sufficient detail to discriminate specific labs and clinical findings. However, data science ultimately seeks to apply computational reasoning and inferencing across data collections and public datasets. Static annotations, while establishing unique identities for biomedical concepts, do not contribute to the goals of reasoning and inference absent asserted relationships between and among a) the concepts within a specific terminology such as LOINC, and ideally b) concepts in related terminologies and ontologies. The core purpose of this proposal is to engineer LOINC content so that datasets that are annotated with LOINC elements (codes and concepts) will facilitate data science analytics. This will be achieved through OWL rendering, linkage to well-formed external ontologies, demonstrating applications that leverage the logical associations, and engaging the LOINC and data science communities to prioritize and validate these efforts. We will restructure LOINC components, terms, and codes into an Ontology Web Language (OWL) rendering to support reasoning. This will include the formalization of LOINC groups and potential related aggregations under “uber codes” (e.g. all blood glucoses). We will link LOINC Components Parts to external, unencumbered ontologies such as Chemical Entities of Biological Interest (ChEBI). These linkages can inform the hierarchy and relationships asserted in the OWL structure. We will demonstrate the application of OWL and related hierarchical reasoning services to allow lumping, splitting and linking of clinical data that is directly or indirectly anchored in LOINC. Using FHIR examples, provide examples and code libraries that allow observations to be queried and aggregated (e.g. all blood glucoses). Reasoning LOINC will be distributed as an open-access resource, in harmony with the OBO community and related biomedical terminology and classification resources. We will leverage existing groups and organizations such as LOINC Users group, CD2H, and ACT, to solicit use cases and dynamically evaluate ontology development and priorities.