Partitioning to Support Auditing and Extending the UMLS

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): The UMLS is an invaluable resource to the biomedical community. However, the Metathesaurus's (META's) size and complexity can hinder its usefulness. We propose a number of divide and conquer techniques that facilitate the tasks of auditing and extending the META. The divide and conquer approach breaks down a large collection of items into more manageable units. In this research, the outcome will be new partitions of the META. Each partition will be employed in auditing the META. The UMLS's Semantic Network (SN) does not partition the META into disjoint sets. We first create a Refined Semantic Network (RSN) that does partition the META into disjoint sets of concepts (extents). Derived from the SN, the RSN incorporates intersection types along with pure semantic types and improves the UMLS in a number of ways, e.g., it contains types with semantically uniform and (usually) smaller extents. In addition to the RSN, three additional, novel partitioning techniques will be applied to extents: Partition along the dimension of relationship similarity; create singly rooted subgroups; and subdivide tangled, multi-rooted hierarchies. Each partition will (normally) expose new errors. We propose a large scale audit of the 2006AA UMLS concentrating on small groups obtained by the partition. The kinds of errors expected to be uncovered include: concept ambiguity and synonymy, wrong and missing IS-A and semantic relationships, inconsistent and incorrect classifications and omissions. We will perform an enhanced audit of the genomic component of the UMLS. For this we will create a Refined Genomic SN (RGSN). We will build an auditing software tool that incorporates the RSN and the three additional levels of partitioning. A formal evaluation, with human subjects, of the tool and a two-level expert evaluation of the auditing results are planned. Finally we focus on the integration of new terminologies into the UMLS, based on the RSN. In summary we will: 1) Create the RSN that partitions META into disjoint units; 2) Design three new partitioning techniques for extents of semantic types; 3) Audit the concepts of small groups; 4) Design a Refined Genomic SN (RGSN) for auditing genomic concepts; 5) Build an auditing tool; 6) Design and evaluate an integration technique for terminologies.