PharmGKB Drug Data Normalization with NDF-RT

Abstract

The Pharmacogenomics Knowledge Base (PharmGKB) [1] is a publicly available central resource for pharmacogenomics data and knowledge, which is being widely employed into clinical practice and thus requires using clinical terminologies. Hence, harmonizing PharmGKB drug data with well annotated drug terminologies will facilitate its integration with other related resources and support data representation, interpretation, and exchange within and across heterogeneous sources and applications. In this study, we extracted drug and drug class from PharmGKB, and mapped them to National Drug File Reference Terminology (NDF-RT) [2], which is integrated into RxNorm specified by U.S. Meaningful Use regulations. And the evaluated mapping results will be provided to PharmGKB for further investigation and collaboration.

Materials and Methods

In this paper we demonstrate our approach to normalize drug data from PharmGKB with NDF-RT.

PharmGKB

PharmGKB contains genomic, phenotype and clinical information collected from ongoing pharmacogenomics studies. We downloaded the drug data on Jan 23th, 2012 snapshot, from which we extracted drug and drug class information.

NDF-RT

NDF-RT drug concepts including VA classifications of medications, generic ingredient preparations used in medications, and orderable (clinical) VA drug products were accessed by NDF-RT web API [3].

Mapping between PharmGKB and NDF-RT

For each drug and drug class, PharmGKB provides generic name, alternate name and brand name that were used for searching NDF-RT identifier (NUI), RxNorm identifier (RxCUI), Unified Medical Language System (UMLS) CUI via NDF-RT API. To retrieve drug class information for an individual drug, we traced back the entire NDF-RT hierarchical tree to record and store a sub tree including information about queried drug and all possible parent nodes (drug class). All the mappings have been evaluated by exact string mapping automatically, non-exact mappings will be manually checked in the further study.

Results

Drug mapping

Total 1,876 drugs and 1,142 drug classes were used in this study. Of these, 1,875 were distinct drugs, and 1,464 were mapped to NDF-RT via 1,563 individual mappings, which included one to many mappings. 1,403 unique PharmGKB drugs were mapped by using generic name, and 61 drugs were mapped by using alternate names. None of the mappings comprised using brand names.

Drug class mapping

Of 1,142 drug classes, 1,082 were distinct drug classes. Only 47 out of 1,082 unique classes were mapped to NDF-RT drug classes resulting into 59 mappings including several one to many mappings. None of the mappings comprised using alternate names and brand names.

Discussion and Conclusion

In this study, we were focusing on drug mapping, about 80% PharmGKB drugs have been successfully mapped to NDF-RT. The reason hinders the rest of mappings failed with NDF-RT are that many PharmGKB drugs and drug classes are named as chemical IUPAC names, and some are chemical compounds used for diagnosis; they are not captured by NDF-RT.

The preliminary normalization work illustrates drugs from different resources represented in site or study specific way by using the local identifier that prevents the data integration and data representation for drug-related clinical studies. Drugs classified by different resources are based on their own perspective. Hence, this normalization work will be benefited across not just biomedical research community, also clinical research side.