The National Institutes of Health's Big Data to Knowledge (BD2K) initiative: capitalizing on biomedical big data

  • Published on

  • View

  • Download


  • The National Institutes of Healths Big Datato Knowledge (BD2K) initiative: capitalizingon biomedical big dataRonald Margolis,1 Leslie Derr,2 Michelle Dunn,3 Michael Huerta,4 Jennie Larkin,5

    Jerry Sheehan,4 Mark Guyer,6 Eric D Green6

    1National Institute of Diabetesand Digestive and KidneyDiseases, NIH, Bethesda,Maryland, USA2Office of the Director, NIH,Bethesda, Maryland, USA3National Cancer Institute,NIH, Bethesda, Maryland, USA4National Library of Medicine,NIH, Bethesda, Maryland, USA5National Heart, Lung andBlood Institute, NIH, Bethesda,Maryland, USA6National Human GenomeResearch Institute, NIH,Bethesda, Maryland, USA

    Correspondence toDr Ronald Margolis, NationalInstitute of Diabetes andDigestive and Kidney Diseases,NIH, 6707 Democracy Blvd,Room 693, Bethesda, MD20892-5460, USA;

    Received 8 May 2014Accepted 17 June 2014

    To cite: Margolis R, Derr L,Dunn M, et al. J Am MedInform Assoc PublishedOnline First: [please includeDay Month Year]doi:10.1136/amiajnl-2014-002974

    ABSTRACTBiomedical research has and will continue to generatelarge amounts of data (termed big data) in manyformats and at all levels. Consequently, there is anincreasing need to better understand and mine the datato further knowledge and foster new discovery. TheNational Institutes of Health (NIH) has initiated a BigData to Knowledge (BD2K) initiative to maximize the useof biomedical big data. BD2K seeks to better define howto extract value from the data, both for the individualinvestigator and the overall research community, createthe analytic tools needed to enhance utility of the data,provide the next generation of trained personnel, anddevelop data science concepts and tools that can bemade available to all stakeholders.

    INTRODUCTION TO THE PROBLEMAcross many areas of science, technological and con-ceptual advances are resulting in the increasinglyrapid generation of large amounts of data. Whileresearch has always involved the collection andorganization of data, the volume, variety, and vel-ocity of current big data production presents newopportunities and challenges in both scale and com-plexity. At the same time, there is a broader culturalshift underway from approaches that kept datamostly private with sharing of resultant knowledgein the form of publications to an information-basedculture that dynamically engages the scientific com-munity through the active sharing of both data andpublications. Big data are not only a new reality forthe biomedical scientist, but an imperative that mustbe understood and used effectively in the quest fornew knowledge. Needed are new approaches fordata management and analysis that allow scientiststo better access and extract value from data so as toadvance research and discovery.1 2 Key stakeholdersin the coming biomedical big data ecosystem includedata providers and users (eg, biomedical researchers,clinicians, and citizens), data scientists, funders, pub-lishers, and libraries. Implementation of such anenvisioned biomedical big data ecosystem willdepend upon cultural changes and require updatedpolicies related to funding, data sharing, and datacitation. Metrics on the efficacy of data sharing andre-use may help inform important decisions atfunding agencies and research institutions about thesupport for and career advancement of biomedicalscientists, and could be invaluable in incentivizingnecessary cultural changes across the biomedicalresearch enterprise. The aim of this perspective is todescribe how the National Institutes of Health

    (NIH) has started catalyzing such changes throughits Big Data to Knowledge (BD2K) initiative.

    BACKGROUNDA report to the NIH Advisory Committee to theDirector from its Data and Informatics WorkingGroup (DIWG) in June 2012 ( recommended the establishment ofa broad and inclusive trans-NIH program foraddressing the opportunities and challenges pre-sented by biomedical big data. Indeed, the spirit ofthe report suggested that failure of the NIH to actat this crucial juncture would represent institu-tional malpractice (as stated by one DIWGmember). To emphasize the need for a coherentand forward-looking response to these opportun-ities, the position of Associate Director for DataScience (ADDS) at NIH was created with the goalof placing big data and data science at the highestlevel of decision-making at the NIH. The firstADDS, Dr Philip Bourne, has begun his tenure atthe NIH, and among his responsibilities is oversightof the BD2K initiative. BD2K has been formulatedas a programmatic response to the DIWG recom-mendations and consists of four focused areas: (1)improving the ability to locate, access, share, anduse biomedical big data; (2) developing and dissem-inating data analysis methods and software; (3)enhancing training in biomedical big data and datascience; and (4) establishing centers of excellencein data science.

    APPROACHTaking into account the substantial current invest-ment of individual NIH Institutes and Centers inbioinformatics and computational biology withintheir own spheres of interest, BD2K has firstfocused on identifying the pressing general butunaddressed needs related to biomedical big data,assessing community expectations, and identifyingcurrent policies and practices related to the produc-tion, handling, and utilization of biomedical bigdata. Outreach to the diverse community of stake-holders mentioned above has been at the forefrontof these efforts through targeted Requests forInformation (RFI), workshops, consultation withleaders in various fields, as well as extensive discus-sions across agencies and within the NIH about therelationship(s) between potential BD2K compo-nents and other ongoing activities ( resulting input strongly ratified the DIWGs

    recommendations that supporting the development of

    Margolis R, et al. J Am Med Inform Assoc 2014;0:12. doi:10.1136/amiajnl-2014-002974 1


    Copyright 2014 by American Medical Informatics Association. on September 25, 2014 - Published by jamia.bmj.comDownloaded from

  • software and applications for analyzing biomedical big data wasonly part of the solution. Further support came from the Holdren/OSTP memo issued in February 2013 ( and directing all agencies of the government to ensure theresults of federally funded scientific research are made broadlyavailable.

    A first step in fostering the emergence of data science as a dis-cipline relevant to biomedicine is the development of solutionsto specific high-need challenges facing the research community.Thus, the first BD2K Funding Opportunity was for investigator-initiated Centers of Excellence in Data Science(RFA-HG-13-009); this called for proposals to test and validatenovel ideas in data science that not only focused on particularchallenges but also had the potential for broad impact.

    The second launched BD2K area aimed to enhance the train-ing of methodologists and practitioners in data science. Skills indemand under the data science umbrella include computerscience, mathematics and statistics, biomedical informatics,biology and medicine, and others, all incorporated as datascience. At the same time, the generation of large amounts ofdata together with the complex questions being posed, requiresinterdisciplinary teams to design the studies and perform thesubsequent data analyses. The BD2K training initiatives seek toseed the development of investigators in all parts of the researchenterprise who are well-trained in data science. Hand-in-handwith training is the need for cultural changes to assure that thecontributions of scientists well-trained in data science are appre-ciated and rewarded, including the provision of appropriatecareer paths with commensurate incentives and rewards.

    A fundamental question for BD2K is how to enable the identifica-tion, access, and citation of (ie, credit for) biomedical data. TheDIWG proposal for federated data catalogs, as distinct from datarepositories, requires descriptions of and pointers to the data.Inherent in data discovery is the need for a sustainable and scalableplan to create and maintain a discovery system that allows researchersto readily find and cite biomedical data. Indeed, sustainability andscalability are two intertwined issues that must be addressed in orderfor the advances made possible by BD2K to have a lasting effect. Anecessary first step has been recognition of the need to assemble andvalidate ideas drawn from the broader scientific community in devel-oping a Data Discovery Index (DDI). The goal of the initial efforts inthis area is to define an effective and efficient mechanism(s) for index-ing that will enable the discovery of relevant, existing datasetsthrough the use of metadata and index terms. The DDI will enableadvanced approaches to search, integrate, and facilitate visualizationof data. Stakeholders will be encouraged to learn from related effortsin other fields, and conduct short-term pilot studies to explore differ-ent ways in which a DDI might be developed and used. Central todevelopment of the DDI will be the ability to link data to associatedpublications to enhance discovery and facilitate better understandingand interpretation of data and associated analyses. Again, cultural andpolicy changes will likely be needed to both enhance data-sharingpolicies and incentivize data sharing.

    Among the difficult challenges is how best to create greater valuefrom the expanding availability and use of electronic health records(EHRs). Clinical data from EHRs, together with individual healthdata captured by various personal devices, offer considerable oppor-tunities for advancing clinical and biomedical research. Data fromclinical sources could very well provide a cost-effective means tostudy different health interventions, to conduct large-scale surveil-lance of disease incidence and progression in real time, to identifypatient cohorts for recruitment into clinical trials, and more.However, unlike most other forms of biomedical research data,

    clinical data are typically captured outside of traditional research set-tings and must then be re-purposed for research use. Doing so raisesimportant issues of consent and protection of patient privacy.Changes in policies and practices are needed to govern researchaccess to clinical data sources and facilitate their use for evidence-based learning in healthcare. Improved approaches to patientconsent and risk-based assessments of clinical data usage forresearch are also high priorities. Improved quality and quantity ofclinical data available for research (eg, by creating shared libraries ofphenotype elements and algorithms and by providing access to keysources of data), as well as new methodologies for analyzing clinicaldata, are all needed for ethical and informed use of these data.

    While a DDI will undoubtedly help to make data findable andcitable, large datasets can also give value beyond their originalpurpose when the data are appropriately annotated so that they canbe used in a meaningful way. Critical to all of these activities will bethe need for formulating, conducting, and maintaining community-based data and metadata standards efforts for reproducibility thatwill: (1) be driven by the clearly identified needs of the community;(2) include a core group of developers with appropriate expertise;(3) engage stakeholders across the community from the outset andin an ongoing way; and (4) evolve based on broad input to becomeincreasingly dynamic and responsive to community needs.

    CONCLUSIONAddressing the challenges associated with biomedical big data mustof necessity engage all parts of the big data ecosystem. While thesechallenges are complex, they are also addressable. BD2K is deploy-ing an integrated plan of action that will tackle numerous aspects ofthe big data challenge, including multiple elements of data science,training, policy, and community behavior. By fostering open com-munication with the stakeholders, BD2K will contribute to thebroader vision for the future of biomedical big data. While the NIHis only one part of the biomedical big data ecosystem, it can provideleadership for convening stakeholders, providing seed funding,developing metrics for success, implementing a working process,establishing and modifying policies, and supporting basic infrastruc-ture that can catalyze long-term solutions for the biomedicalresearch community. Through a working partnership between andamong the relevant stakeholders, useful solutions will emerge andlead to vibrant and sustainable models for capitalizing on biomed-ical big data and translating data into new knowledge.

    Acknowledgements Helpful edits were provided by Susan Gregorick (NIGMS),Belinda Seto (NIBIB), and Lynda Hardy (NINR).

    Contributors RM conceived of the idea for the article, and performed the literaturesearch. RM and JL jointly wrote the article, with contributions from all other authors.RM and JL share joint responsibility as guarantors.

    Competing interests None.

    Provenance and peer review Not commissioned; externally peer reviewed.

    Data sharing statement The NIH has a full data sharing policy. All publicationsmust be openly available through the NLM.

    Open Access This is an Open Access article distributed in accordance with theCreative Commons Attribution Non Commercial (CC BY-NC 4.0) license, whichpermits others to distribute, remix, adapt, build upon this work non-commercially,and license their derivative works on different terms, provided the original work isproperly cited and the use is non-commercial. See:

    REFERENCES1 Science. Dealing with data. Science 2011;331:639806.

    content/331/6018.toc#SpecialIssue2 Nature. Science in the petabyte era. Nature 2008;455:1136. http://www.nature.


    2 Margolis R, et al. J Am Med Inform Assoc 2014;0:12. doi:10.1136/amiajnl-2014-002974

    Perspective on September 25, 2014 - Published by jamia.bmj.comDownloaded from

  • doi: 10.1136/amiajnl-2014-002974 published online July 9, 2014J Am Med Inform Assoc

    Ronald Margolis, Leslie Derr, Michelle Dunn, et al. on biomedical big datato Knowledge (BD2K) initiative: capitalizing The National Institutes of Health's Big Data information and services can be found at:

    These include:

    Open Access

    non-commercial. See:, provided the original work is properly cited and the use iswork non-commercially, and license their derivative works on different license, which permits others to distribute, remix, adapt, build upon thisCreative Commons Attribution Non Commercial (CC BY-NC 4.0) This is an Open Access article distributed in accordance with the


  • Notes

    (DOIs) and date of initial publication. publication. Citations to Advance online articles must include the digital object identifier citable and establish publication priority; they are indexed by PubMed from initialtypeset, but have not not yet appeared in the paper journal. Advance online articles are Advance online articles have been peer reviewed, accepted for publication, edited and request permissions go to: order reprints go to: subscribe to BMJ go to: on September 25, 2014 - Published by jamia.bmj.comDownloaded from

    The National Institutes of Health's Big Data to Knowledge (BD2K) initiative: capitalizing on biomedical big dataAbstractIntroduction to the problemBackgroundApproachConclusionReferences


View more >