Data at the NIH: Some Early Thoughts

Data at the NIH: Some Early Thoughts
Philip E. Bourne Ph.D.
Associate Director for Data Science
National Institutes of Health
http://www.slideshare.net/pebourne/

Background
 Research in computational biology…
 Co-directed the RCSB Protein Data Bank (1999-
2014)
 Co-founded PLOS Computational Biology; First EIC
(2005 – 2012)
 With Ontologies:
– Extensive work with the Gene Ontology
– Co-developed mmCIF for macromolecular structure

Disclaimer: I only started March 3,
2014
…but I had been thinking about this prior to my
appointment
http://pebourne.wordpress.com/2013/12/

Numberofreleasedentries
Year
Motivation for Change:
PDB Growth in Numbers and Complexity
[From the RCSB Protein Data Bank]

Motivation for Change:
We Are at the Beginning

Motivation:
We Are at an Inflexion Point for Change
 Evidence:
– Google car
– 3D printers
– Waze
– Robotics

From the Second Machine Age
From: The Second Machine Age: Work, Progress, and
Prosperity in a Time of Brilliant Technologies by Erik
Brynjolfsson & Andrew McAfee

Much Useful Groundwork Has Been
Done

NIH Data & Informatics WorkingNIH Data & Informatics Working
GroupGroup
In response to the growth of large biomedicalgrowth of large biomedical
datasetsdatasets, the Director of NIH established a
special Data and Informatics Working Group
(DIWG).

Big Data to Knowledge (BD2K)Big Data to Knowledge (BD2K)
1. Facilitating Broad Use
2. Developing and Disseminating Analysis
Methods and Software
3. Enhancing Training
4. Establishing Centers of Excellence
http://bd2k.nih.gov

Currently…
 Data Discovery Index – under review
 Data Centers – under review
 Training grants – RFA’s issued; under review
 Software index – workshop in May
 Catalog of standards – FOA under development

This is just the beginning…
Some Early Observations

1. We don’t know enough about how existing data are
used

* http://www.cdc.gov/h1n1flu/estimates/April_March_13.htm
Jan. 2008 Jan. 2009 Jan. 2010Jul. 2009Jul. 2008 Jul. 2010
1RUZ: 1918 H1 Hemagglutinin
Structure Summary page activity for
H1N1 Influenza related structures
3B7E: Neuraminidase of A/Brevig Mission/1/1918
H1N1 strain in complex with zanamivir
[Andreas Prlic]
Consider What
Might Be Possible

We Need to Learn from Industries Whose
Livelihood Addresses the Question of Use

used
2. We have focused on the why, but not the how

2. We have focused on the why, but
not the how
 The OSTP directive is the why
 The how is needed for:
– Any data that does not fit the existing data resource model
• Data generated by NIH cores
• Data accompanying publications
• Data associated with the long tail of science

Considering a Data Commons to
Address this Need
 AKA NIH drive – a dropbox for NIH investigators
 Support for provenance and access control
 Likely in the cloud
 Support for validation of specific data types
 Support for mining of collective intramural and
extramural data across IC’s
 Needs to have an associated business model
http://100plus.com/wp-content/uploads/Data-Commons-3-1024x825.png

used
3. We do not have an NIH-wide sustainability plan for
data (not heard of an IC-based plan either)

3. Sustainability
 Problems
– Maintaining a work force – lack of reward
– Too much data; too few dollars
– Resources
• In different stages of maturity but treated the same
• Funded by a few used by many
– True as measured by IC
– True as measured by agency
– True as measured by country
• Reviews can be problematic

3. Sustainability
 Possible Solutions
– Establish a central fund to support
– The 50% model
– New funding models eg open submission and review
– Split innovation from core support and review separately
– Policies for uniform metric reporting
– Discuss with the private sector possible funding models
– More cooperation, less redundancy across agencies
– Bring foundations into the discussion
– Discuss with libraries, repositories their role
– Educate decision makes as to the changing landscape

used
4. Training in biomedical data science is spotty

4. Training in biomedical data science
is spotty
 Problem
– Coverage of the domain is unclear
– There may well be redundancies
 Solution
– Cold Spring Harbor like training facility(s)
• Training coordinator
• Rolling hands on courses in key areas
• Appropriate materials on-line
– Interagency training initiatives

used
4. Training in biomedical data science is spotty
5. Reproducibility will need to be embraced

47/53 “landmark” publications
could not be replicated
[Begley, Ellis Nature,
483, 2012] [Carole Goble]

I can’t reproduce research from my
own laboratory?
Daniel Garijo et al. 2013 Quantifying Reproducibility in Computational Biology:
The Case of the Tuberculosis Drugome PLOS ONE 8(11) e80278 .

Characteristics of the Original and
Current Experiment
 Original and Current:
– Purely in silico
– Uses a combination of public databases and
open source software by us and others
 Original:
– http://funsite.sdsc.edu/drugome/TB/
 Current:
– Recast in the Wings workflow system
Daniel Garijo et al. 2013 Quantifying Reproducibility in Computational Biology:
The Case of the Tuberculosis Drugome PLOS ONE 8(11) e80278 .

Considered the Ability to Reproduce
by Four Classes of User
 REP-AUTHOR – original author of the work
 REP-EXPERT – domain expert – can reproduce even
with incomplete methods described
 REP-NOVICE – basic domain (bioinformatics) expertise
 REP-MINIMAL – researcher with no domain expertise
Garijo et al 2013 PLOS ONE 8(11): e80278

A Conceptual Overview of the Method
Should Be Mandatory

Time to Reproduce the Method

Its not that we could not reproduce
the work, but the effort involved was
substantial
Any graduate student could tell you
this and little has changed in 40 years
Perhaps it is time we did better?

I cast the solutions in a vision …
something I call the digital enterprise
Any institution is a candidate to be a digital
enterprise, but lets explore it in the context
of the academic medical center

Components of The Academic Digital
Enterprise
 Consists of digital assets
– E.g. datasets, papers, software, lab notes
 Each asset is uniquely identified and has provenance,
including access control
– E.g. publishing simply involves changing the access control
 Digital assets are interoperable across the enterprise

Life in the Academic Digital Enterprise
 Jane scores extremely well in parts of her graduate on-line neurology class.
Neurology professors, whose research profiles are on-line and well described, are
automatically notified of Jane’s potential based on a computer analysis of her scores
against the background interests of the neuroscience professors. Consequently,
professor Smith interviews Jane and offers her a research rotation. During the
rotation she enters details of her experiments related to understanding a widespread
neurodegenerative disease in an on-line laboratory notebook kept in a shared on-line
research space – an institutional resource where stakeholders provide metadata,
including access rights and provenance beyond that available in a commercial
offering. According to Jane’s preferences, the underlying computer system may
automatically bring to Jane’s attention Jack, a graduate student in the chemistry
department whose notebook reveals he is working on using bacteria for purposes of
toxic waste cleanup. Why the connection? They reference the same gene a number
of times in their notes, which is of interest to two very different disciplines – neurology
and environmental sciences. In the analog academic health center they would never
have discovered each other, but thanks to the Digital Enterprise, pooled knowledge
can lead to a distinct advantage. The collaboration results in the discovery of a
homologous human gene product as a putative target in treating the
neurodegenerative disorder. A new chemical entity is developed and patented.
Accordingly, by automatically matching details of the innovation with biotech
companies worldwide that might have potential interest, a licensee is found. The
licensee hires Jack to continue working on the project. Jane joins Joe’s laboratory,
and he hires another student using the revenue from the license. The research
continues and leads to a federal grant award. The students are employed, further
research is supported and in time societal benefit arises from the technology.
From What Big Data Means to Me JAMIA 2014 21:194

Life in the NIH Digital Enterprise
 Researcher x is made aware of researcher y through
commonalities in their data located in the data commons.
Researcher x reviews the grants profile of researcher y and
publication history and impact from those grants in the past 5
years and decides to contact her. A fruitful collaboration ensues
and they generate papers, data sets and software. Metrics
automatically pushed to company z for all relevant NIH data and
software in a specific domain with utilization above a threshold
indicate that their data and software are heavily utilized and
respected by the community. An open source version remains,
but the company adds services on top of the software for the
novice user and revenue flows back to the labs of researchers x
and y which is used to develop new innovative software for
open distribution. Researchers x and y come to the NIH training
center periodically to provide hands-on advice in the use of their
new version and their course is offered as a MOOC.

To get to that end point we have to
consider the complete research
lifecycle

The Research Life Cycle will Persist
IDEAS – HYPOTHESES – EXPERIMENTS – DATA - ANALYSIS - COMPREHENSION - DISSEMINATION

Tools and Resources Will Continue To
Be Developed
Authoring
Tools
Lab
Notebooks
Data
Capture
Software
Analysis
Tools
Visualization
Scholarly
Communication

Those Elements of the Research Life Cycle will
Become More Interconnected Around a Common
Framework
Authoring
Tools
Lab
Notebooks
Data
Capture
Software
Analysis
Tools
Visualization
Scholarly
Communication

New/Extended Support Structures Will
Emerge
Authoring
Tools
Lab
Notebooks
Data
Capture
Software
Analysis
Tools
Visualization
Scholarly
Communication
Commercial &
Public Tools
Git-like
Resources
By Discipline
Data Journals
Discipline-
Based Metadata
Standards
Community Portals
Institutional Repositories
New Reward
Systems
Commercial Repositories
Training

We Have a Ways to Go
Authoring
Tools
Lab
Notebooks
Data
Capture
Software
Analysis
Tools
Visualization
Scholarly
Communication
Commercial &
Public Tools
Git-like
Resources
By Discipline
Data Journals
Discipline-
Based Metadata
Standards
Community Portals
Institutional Repositories
New Reward
Systems
Commercial Repositories
Training

Next Steps
 Support for research objects
– These objects underpin the various cataloging efforts
 Support for data metrics
– Such metrics underpin a change in the reward system

NIHNIH……
Turning Discovery Into HealthTurning Discovery Into Health
philip.bourne@nih.gov

Thank You!
Questions?
philip.bourne@nih.gov

Back Pocket Slides for BD2K
Programs

1. Facilitating Broad Use1. Facilitating Broad Use

1. Facilitating Broad Use1. Facilitating Broad Use
 Research use of clinical data
 Workshop held Sept 2013
 Workshop report and plans being finalized
 Contacts: Jerry Sheehan (NLM) and Leslie Derr (OD)
 Community-based data and metadata
standards
 Will make data usable
 Workshop held Sept 2013
 Workshop report and plans being finalized
 Contact: Mike Huerta (NLM)

2. Facilitating Big Data Analysis2. Facilitating Big Data Analysis
 Broad-based, on-going BISTI PARs
– BISTI: Biomedical Information Science and Technology
Initiative
– Joint BISTI-BD2K effort
– R01s and SBIRs
– Contacts: Peter Lyster (NIGMS) and Jennifer Couch (NCI)
 Planned Workshops:
– Software Index (Spring 2014)
• Need to be able to find and cite software, as well as data, to
support reproducible science.
– Cloud Computing (Summer/Fall 2014)
• Biomedical big data are becoming too large to be analyzed on
traditional localized computing systems.
– Contact: Vivien Bonazzi (NHGRI)

2. Facilitating Big Data Analysis2. Facilitating Big Data Analysis
 RFA for Targeted Software Development
Development of Software and Analysis Methods
for Biomedical Big Data in Targeted Areas of
High Need (U01)
–RFA-HG-14-020
–Application receipt date June 20, 2014
–Topics: data compression/reduction, visualization,
provenance, or wrangling.
–Contact: Jennifer Couch (NCI) and Dave Miller (NCI)
http://bd2k.nih.gov

3: Enhancing Training3: Enhancing Training
 Summary of Training Workshop and Request for
Information:
– http://bd2k.nih.gov/faqs_trainingFOA.html
– Contact: Michelle Dunn (NCI)
 Training Goals:
– develop a sufficient cadre of researchers skilled in the
science of Big Data
– elevate general competencies in data usage and analysis
across the biomedical research workforce.

3: BD2K Training RFAs3: BD2K Training RFAs
Application Receipt Date: April 2, 2014
K01s for Mentored Career Development Awards,
RFA-HG-14-007
 Provides salary and research support for 3-5 years for
intensive research career development under the guidance
of an experienced mentor in biomedical Big Data Science.
R25s for Courses for Skills Development, RFA-HG-
14-008
 Development of creative educational activities with a
primary focus on Courses for Skills Development.
R25 for Open Educational Resources, RFA-HG-14-
009
 Development of open educational resources (OER) for use
by large numbers of learners at all career levels, with a
primary focus on Curriculum or Methods Development.

4:4: BD2K Centers of ExcellenceBD2K Centers of Excellence
 Two or more rounds of center awards
 FY14
 Investigator-initiated Centers of Excellence for Big
Data Computing in the Biomedical Sciences (U54)
RFA-HG-13-009 (closed)
 BD2K-LINCS-Perturbation Data Coordination and
Integration Center (DCIC) (U54) RFA-HG-14-001
(closed)

Data at the NIH: Some Early Thoughts

More Related Content

What's hot (20)

Similar to Data at the NIH: Some Early Thoughts (20)

More from Philip Bourne (20)

Recently uploaded (20)

Data at the NIH: Some Early Thoughts

Editor's Notes