background image
Odds ratios of studies with links from GEO or ArrayExpress
0.0
1.0
2.0
3.0
4.0
5.0
6.0
Funded by 3 sources vs none
Funded by 2 sources vs none
Funded by 1 source vs none
Funded by Non-US Govt
Funded by US Govt Non-NIH
Funded by NIH
Humans w hen cancer=FALSE
Humans w hen cancer=TRUE
Cancer
Humans
Published in a core clinical journal
Published w ith free full text
Published in 2006 vs 2002
Odds Ratio
Prevalence and Patterns
of Microarray Data Sharing
Heather A. Piwowar, Wendy W. Chapman
Background
· Sharing research data is
a cornerstone of science
· Prevalence of data sharing is not well
understood
· The most comprehensive method for
measuring occurrences of public data
sharing is manual curation
· Our preliminary manual curation of 100
papers:
· 30% of investigators publicly share
their full microarray datasets:
· 70% in NCBI's Gene Expression
Omnibus (GEO)
· 20% at EBI's ArrayExpress
· 10% in smaller databases or lab
or publisher websites.
· Manual curation is extremely time
consuming
· Here we perform larger-scale
automated estimates
Limitations
· Prevalence depends on study
selection criteria
· Some of the retrieved papers only
reuse data, and thus are not relevant to
the question of sharing. This may
confound our results, but we estimate
the effect is small.
Conclusions
· Natural language processing
techniques could be helpful for curation
based on full-text data sharing
statements
· Our automated approach yielded
conservative estimates of 20% sharing
within GEO and ArrayExpress
· Prevalence patterns suggest several
promising hypotheses for understanding
data sharing behavior
· We hope these preliminary results will
inspire additional investigations into data
sharing behavior, and in turn the
development of effective policies and
tools to facilitate this important aspect of
scientific research.
Acknowledgments
HAP funded by NLM Training Grant
For more information: heather.piwowar@alumni.pitt.edu
Poster will be available at Nature Precedings.
University of Pittsburgh
Department of Biomedical Informatics
About 20% of microarray studies had links
from publicly available databases
· Extracted research articles with MeSH terms for
both "Gene Expression Profiling" and
"Oligonucleotide Array Sequence Analysis"
published in 2006
· Searched GEO and ArrayExpress for links to
these PubMed IDs to determine which articles had
been credited as an originating data source
· 2503 articles
· 440 (18%) had links from either GEO or
ArrayExpress:
· 70% had links from GEO
· 30% had links from ArrayExpress
· Overlapping 12% from both
What proportion of microarray datasets are shared in public databases?
Prevalence varies with time, publishing decisions, study topic and subjects, and funding source
· Prevalence of database links is 2.5x higher in 2006 than 2002.
· Studies with free full text at PubMed are 2.1x more likely to have links
· Studies published in a core clinical journal are 1.4x likely to have links as those published elsewhere
· Studies with human data were less likely to have a link when the trial was related to cancer (OR=0.8), but otherwise no
different than studies on other species
· Studies funded by NIH were 2.0x more likely to have links than studies not funded by the NIH
· Increased number of funding sources correlates with a higher prevalence of sharing
Does sharing data correlate with other factors?
Proportion of m icroarray article s w ith links from database s
0
10
20
30
40
50
60
70
80
90
100
All artic les (n= 2503)
Artic les w ith links from databas es (n=440)
P
e
r
c
e
n
t

o
f

t
o
t
a
l
No links from
databases
(implying data not
deposited in
GEO or
ArrayExpress)
Links from
databases
Links from GEO
Links from
ArrayExpress
Links from both
Nature Precedings : doi:10.1038/npre.2008.1701.1 : Posted 20 Mar 2008