The demands of data-intensive science represent a challenge for diverse scientific communities. Data volumes from various sources are increasing exponentially, creating data management challenges. New approaches and technologies are needed to enable scientists to effectively analyze and store massive amounts of data.

1. Data-Intensive Research
Jano van Hemert
research.nesc.ac.uk
NI VER
U S
E
IT
TH
Y
O F
H
G
E
R
D I U
N B

2. Downloaded from www.sciencemag.org on July 6, 2009
COMPUTER SCIENCE
The demands of data-intensive science
Beyond the Data Deluge represent a challenge for diverse scientific
communities.
Gordon Bell,1 Tony Hey,1 Alex Szalay2
S
ince at least Newton’s laws of motion in
the 17th century, scientists have recog-
nized experimental and theoretical sci-
ence as the basic research paradigms for
understanding nature. In recent decades, com-
puter simulations have become an essential
third paradigm: a standard tool for scientists to
explore domains that are inaccessible to theory
and experiment, such as the evolution of the
universe, car passenger crash testing, and pre-
dicting climate change. As simulations and
experiments yield ever more data, a fourth par-
adigm is emerging, consisting of the tech-
niques and technologies needed to perform
data-intensive science (1). For example, new
types of computer clusters are emerging that
are optimized for data movement and analysis
rather than computing, while in astronomy and
other sciences, integrated data systems allow
data analysis and storage on site instead of
requiring download of large amounts of data. Moon and Pleiades from the VO. Astronomy has been one of the first disciplines to embrace data-intensive
Today, some areas of science are facing science with the Virtual Observatory (VO), enabling highly efficient access to data and analysis tools at a cen-
hundred- to thousandfold increases in data tralized site. The image shows the Pleiades star cluster form the Digitized Sky Survey combined with an image
volumes from satellites, telescopes, high- of the moon, synthesized within the World Wide Telescope service.
throughput instruments, sensor networks,
accelerators, and supercomputers, compared challenging scientists (4). In contrast to the tra- ing of these digital data are becoming increas-
to the volumes generated only a decade ago ditional hypothesis-led approach to biology, ingly burdensome for research scientists.
(2). In astronomy and particle physics, Venter and others have argued that a data- Over the past 40 years or more, Moore’s
these new experiments generate petabytes intensive inductive approach to genomics Law has enabled transistors on silicon chips to
CREDIT: JONATHAN FAY/MICROSOFT
(1 petabyte = 1015 bytes) of data per year. In (such as shotgun sequencing) is necessary to get smaller and processors to get faster. At the
bioinformatics, the increasing volume (3) and address large-scale ecosystem questions (5, 6). same time, technology improvements for
the extreme heterogeneity of the data are Other research fields also face major data disks for storage cannot keep up with the ever
management challenges. In almost every labo- increasing flood of scientific data generated
ratory, “born digital” data proliferate in files, by the faster computers. In university research
1MicrosoftResearch, One Microsoft Way, Redmond, WA spreadsheets, or databases stored on hard labs, Beowulf clusters—groups of usually
98052, USA. 2Department of Physics and Astronomy, Johns
Hopkins University, 3701 San Martin Drive, Baltimore, MD drives, digital notebooks, Web sites, blogs, and identical, inexpensive PC computers that can
21218, USA. E-mail: szalay@jhu.edu wikis. The management, curation, and archiv- be used for parallel computations—have
www.sciencemag.org SCIENCE VOL 323 6 MARCH 2009 1297
Published by AAAS

3. o investigate the 10.1126/science.1171406
Downloaded from www.sciencemag.org on July 6, 2009
COMPUTER SCIENCE
The demands of data-intensive science
Beyond the Data Deluge represent a challenge for diverse scientific
communities.
Gordon Bell,1 Tony Hey,1 Alex Szalay2
S
ince at least Newton’s laws of motion in
the 17th century, scientists have recog-
nized experimental and theoretical sci-
The demands of data-intensive science
ence as the basic research paradigms for
understanding nature. In recent decades, com-
puter simulations have become an essential
represent a challenge for diverse scientific
third paradigm: a standard tool for scientists to
explore domains that are inaccessible to theory
and experiment, such as the evolution of the
communities.
universe, car passenger crash testing, and pre-
dicting climate change. As simulations and
experiments yield ever more data, a fourth par-
adigm is emerging, consisting of the tech-
niques and technologies needed to perform
data-intensive science (1). For example, new
types of computer clusters are emerging that
are optimized for data movement and analysis
rather than computing, while in astronomy and
other sciences, integrated data systems allow
data analysis and storage on site instead of
requiring download of large amounts of data. Moon and Pleiades from the VO. Astronomy has been one of the first disciplines to embrace data-intensive
Today, some areas of science are facing science with the Virtual Observatory (VO), enabling highly efficient access to data and analysis tools at a cen-
hundred- to thousandfold increases in data tralized site. The image shows the Pleiades star cluster form the Digitized Sky Survey combined with an image
volumes from satellites, telescopes, high- of the moon, synthesized within the World Wide Telescope service.
throughput instruments, sensor networks,
accelerators, and supercomputers, compared challenging scientists (4). In contrast to the tra- ing of these digital data are becoming increas-
to the volumes generated only a decade ago ditional hypothesis-led approach to biology, ingly burdensome for research scientists.
(2). In astronomy and particle physics, Venter and others have argued that a data- Over the past 40 years or more, Moore’s
these new experiments generate petabytes intensive inductive approach to genomics Law has enabled transistors on silicon chips to
CREDIT: JONATHAN FAY/MICROSOFT
(1 petabyte = 1015 bytes) of data per year. In (such as shotgun sequencing) is necessary to get smaller and processors to get faster. At the
bioinformatics, the increasing volume (3) and address large-scale ecosystem questions (5, 6). same time, technology improvements for
the extreme heterogeneity of the data are Other research fields also face major data disks for storage cannot keep up with the ever
management challenges. In almost every labo- increasing flood of scientific data generated
ratory, “born digital” data proliferate in files, by the faster computers. In university research
1MicrosoftResearch, One Microsoft Way, Redmond, WA spreadsheets, or databases stored on hard labs, Beowulf clusters—groups of usually
98052, USA. 2Department of Physics and Astronomy, Johns
Hopkins University, 3701 San Martin Drive, Baltimore, MD drives, digital notebooks, Web sites, blogs, and identical, inexpensive PC computers that can
21218, USA. E-mail: szalay@jhu.edu wikis. The management, curation, and archiv- be used for parallel computations—have
www.sciencemag.org SCIENCE VOL 323 6 MARCH 2009 1297
Published by AAAS

4. NEWS FEATURE 2020 COMPUTING NATURE|Vol 440|23 March 2006
J. MAGEE
EVERYTHING,EVERYWHERE
Tiny computers that constantly monitor ecosystems, buildings and even human bodies
could turn science on its head. Declan Butler investigates.

5. o P,;,(>.?.;90:(,1;.=(/(7.,=<G(
- 40(J<.(;5.(!%@)(10=(=.<.,=D5(A0J?>(=.IJ9=.B( #;'#*"(2) !"2:/"1#("-,0+"2*0+
1#("2*0+!/(04")2(
)581,7-56@.2.35=52/
3000 (7>.26!/,+058
#;'#*'+.'(2 &0(.3"!"#//+"50(
)0'.3"!"#//+"50(
*#&"!"50(2)."(2)
82)."!"#//+"50(
o 45,;(;5.(=.IJ9=.F.:;<(A.=.(>.19:.>G(
36$(0-,0+04,# 50(2).")*,
!"5,)3"-,0+
!"50(2).
#&&"#))0*"#1"4204(
(24").'$
*#&"!"5,)3"#;'#.")*,
5(2)38#.2("-,0+ 50(2)."2*0+"1#&#4
,&."!"'(-#&"(24,0&#+
30'),&4").'$ !")/2*"2$'*
,&.2(%")*3"*+,&
.","-(,."4204(
/(04"3'1"4204 -(,."!")0*,0+"2$'*
(21")/2*"2$'*
)5=1/56!528023 #&&"4+#*,0+ #&.,/0$2
'(-#&").'$
$+,-./012 G01;139
o 45.()D0;;9<5(Q0C.=:F.:;(,:>(L.,?;5(<.=C9D.(8=0C9>.=<(A.=.(8.=<J,>.>(;5,;(9;(
!"4+#*,0+
,&."!"(210.2")2&)
(210.2")2&)"2&%,(0& !"+2#(&"$,)#-,+ 10+"2*0+
-(,."2$'*"(2)"! !"#//+"2*0+
/(04"/36)"4204 0,70)
-23#%,0'( -,0+"*0&)2(%
#1"!"/)6*3,#.
Disk space (Terabytes)
-23#%"2*0+
2500
02*0+04,#
A,<(9F80=;,:;(,:>(:.D.<<,=7G(
#&,1"-23#%
!"0**'/"0(4#&"/)6*3 *0&)2(%"-,0+
30+0*2&2
/2(*2/."/)6*30/36)
-(,."!"2$'*"/)6*30+ %5,718-052-5 &2'(0,1#42
2*0+046 2%0+'.,0&
1210(6
)2:"(0+2) ;"!"2:/"/)6*30+!# !"*04&,.,%2"&2'(0)*,
/)6*30&"-"(2%
!"1#((,#42"5#1
121"*04&,.,0& /36),*#"* @5+0-025
<51;139 /)6*30+")*,
+#&*2.
&28"2&4+"!"12$
o 45.(D0<;(01(<;0=,-.(A0J?>(:..>(;0(H.(:.-0;9,;.>G(
0/."+2..
!"420+")0*"+0&$0& ,&.2++,42&*2
-(,."!"/)6*30+
)0*"$2%
"1320/0456!-052-5 !"&2'(0)*,
2000 /"420+04,)."#))0*
/2.(0+"420)*,
-(,."!"$2%"/)6*30+
-(,."!")0*"/)6*30+
*3,+$"$2%
,222"."1#4&
!"%#*")*,".2*3&0+"-
G01-:5=08/79
*+#6"1,&2( !"*3,+$"/)6*30+"/)6*
/2()",&$,%"$,552( ,&."!"+#&4"*011"$,) )0+,$").#.2"*011'& %&'()$
!"/2()")0*"/)6*30+
!"#'.,)1"$2%"$,)0($
*:980-8 /36),*#"-
!"1#4&"1#4&"1#.2(
*2++
!",11'&0+
*+#6"*+#6"1,&2( /36)"(2%"+2..
o 45.(<.=C9D.(A0J?>(:..>(;0(H.(:.-0;9,;.>G( !"#$%"$
/36)"(2%"-
#//+"/36)"+2.. !"%,(0+
#/3#),0+046 !"/36)!*0&$2&)"1#.
!#/#&2)2"!0'(&#+"05"#//+,2$"/36),*)
!"#//+"/36) 2#(.3"#&$"/+#&2.#(6")*,2&*2"+2..2()
!"*(6)."4(08.3
*89-:1;139 @./570.;6!-052-5 )'(5"*0#.".2*3
!"2'("*2(#1")0*
1#.2(".(#&)
#//+"/36)"#!1#.2(
#//+")'(5")*,
!"&0&!*(6).")0+,$)!"#//+"2+2*.(0*321
,&."!"36$(042&"2&2(4 &#&0"+2..
/36)"(2%"2
2+2*.(0*321")0+,$").
!"2+2*.(0*321")0*
!"#++06"*01/$ )'(5")*, 2+2*.(0*321"*011'&
420/36)"(2)"+2..
-,0*321,).(6 -,0*321"-,0/3"(2)"*0
!"10+"-,0+
52-)"+2..
*3,+$"*#(2"32#+.3"#&$"$2%2+0/12&.
-(,."!"*+,&"/)6*30+
!",&.2++"$,)#-,+"(2)
-(,."!"32#+.3"/)6*3
!"#1"*2(#1")0*
1#.2("+2.. !")0+,$").#.2"*321
.3,&")0+,$"5,+1) *321"/36)"+2..
!"*321"/36)
&#&0.2*3&0+046
*%&! &'*+2,*"#*,$)"(2)
)0+,$").#.2",0&,*) -,0/36)"!
1#.")*,"2&4"#!).('*.
#*.#"1#.2( /36)"*321"*321"/36)
!"1#.2("/(0*2))".2*3 )*(,/.#"1#.2( )6&.32.,*"12.
2+2*.(0*3,1"#*.#
*#(-0& !"-,0+"*321
!"/082(")0'(*2)
*321"1#.2(
!"/36)"*321"-
o R.7( 8.08?.( A.=.( ,H?.( ;0( 9:1?J.:D.( ;5.( 8=0>JD;( >.<9-:( ;0( .:<J=.( ;5,;( 9;( F.;(
&2'(0/)6*30+"(23#-,+ !"1#.2(")*,
1500 +#&41',( )0+"2&2(4"1#.")0+"*
32#+.3"9")0*,#+"*#(2",&".32"*011'&,.6 -(#,&",&!'(6 !"1#.2("*321
!"/36)"*321"#
1,*(0/0("12)0/0("1#.
*#.#+"+2.. 420*3,1"*0)10*3,1"#*
!"#1"*321")0* !"420/36)"(2)
-(,."!")0*"80(7 *01/0)")*,".2*3&0+
*,>;0-6?5.;/: #')."&<"!"/)6*3,#.
12.#++"1#.2(".(#&)"# *321"(2%
)0*,0+"32#+.3",++$,)#-,+"(23#-,+
!"&'()")*30+#()3,/ !"(23#-,+"12$ !"1#.2("2&4"/2(50(1
-(,."*2(#1".
!"&#&0)*,"&#&0.2*3&0 1#*(010+2*'+2)!"*#.#+
-,01#*(010+2*'+2)
*0++0,$")'(5#*2"#
*321"*011'&
*321!2'("!
/0+61"$24(#$").#-,+
/0+612( ,&0(4"*321
":5=08/79
*#.#+".0$#6
#//+"*#.#+"#!42&
)*#&$"!"*#(,&4")*, !"/0+61")*,"/0+"*321 !"10+"*#.#+"#!*321
#&#2).32),# /08$2("12.#++ 1#*(010+"(#/,$"*011
/'-+,*"32#+.3"&'() ,&."!"32#."1#))".(#&
&'()"2.3,*) )'(5"2&4 2'("/0+61"!
1#*(010+")61/ 0(4"+2..0(4#&012.#++,*)
#*.#"/)6*3,#.")*#&$
/#++,#.,%2"12$ !"*+,&"&'() ,(0&1#7").22+1#7 !"#//+"/0+61")*, 2'("!"0(4"*321
/0+61",&. 1#*(010+"*321"/36) !"*0++0,$",&.2(5")*,
0(4"-,010+"*321
!"0(4#&012."*321
,&0(4"*3,1"#*.# #//+"*#.#+"-!2&%,(0&
!"#$%"&'() -,00(4#&"12$"*321 *321"2&4")*,
=.IJ9=.F.:;<G(
2'("!"*#&*2("*#(2 12$"2$'* !"0(4"*321
.2.(#32$(0&!#)6112.(
#$%#&*2)",&").('*.'(#+"2&4,&22(,&4
#42"#42,&4 )*,".2*3&0+"82+$"!0, /0+61"2&4")*, )2&)0("#*.'#."-!*321
!"/0+61")*,"/0+"/36)!"12$"*321
-(,."!"42&"/(#*. *+,&"(23#-,+ -,01#.2(,#+) -,00(4"12$"*321"+2..
)6&+2.. )6&.32),)!).'..4#(.
!".2+212$".2+2*#(2 2&2(4"2:/+0("2:/+0,.
1#.2(")*,".2*3")2( .2.(#32$(0&
.2.(#32$(0&"+2.. ,&$"2&4"*321"(2) *01/'."*321"2&4
!"12$")*(22& #,*32"!
!"121-(#&2")*,
?5.;/:6".75 +#-"#&,1
,&."!").$"#,$)
-(,."12$"-'++ /+#)."('--2("*01/0)
!"-,01#.")*,!/0+61"2
!"-,012$"1#.2("(2) !"-#*.2(,0+ 5+',$"/3#)2"2;',+,-(
!"*321"2&4"$#.#
!"36$(0+
8#.2("(2)0'("(2)
*0((0)"2&4")*,".2*3&
#&&".(0/"12$"/#(#),.
,&.2($,)*,/+")*,"(2%
!"#1"42(,#.(")0* #&&"*+,&"-,0*321
*:.7=.-1;139 #.10)"2&%,(0&
1000
!"&#."/(0$'*.)
#&&".(0/"/#2$,#.( )6&.32.,*"*011'&
#&#+"*321 )2/"/'(,5".2*3&0+ 2&2(4"5'2+
/+#&."/36),0+
2'("!"0/2("(2)
$2(1#.0+")'(4
,&."!"/(0$"(2) #$$,*.,0&
,&."!".'-2(*"+'&4"$ &2'(0+"(2)
(2$0:"(2/
!"*0&.(0+"(2+2#)2
,&."!"/3#(1
-,0)2&)"-,02+2*.(0& 2&%,(0&")*,".2*3&0+
$2)#+,&#.,0&
!"3#<#($"1#.2(
$240712=52/.;
,&."!"/(0$"2*0&
$('4"#4,&4 /(2%"12$
!"/3#(1")*,
/36.0*321,).(6
.#+#&.# 2&%,(0&"/0++'.
!-052-5
8#.2("#,("#&$")0,+"/0++'.,0&
!"3'1"&'.("$,2. !"#&#+".0:,*0+
1'+.")*+2( 5'2+ 2&2(4"*0&%2()"1#&#42
A57=./1;139 ,&."!"$2(1#.0+
-(,."!"$2(1#.0+
!"2'("#*#$"$2(1#.0+
*&)"$('4)
!"/3#(1"/3#(1#*0+
/3#(1#*2'."(2)
$('4"$2%",&$"/3#(1
2'("!"/3#(1"-,0/3#(1
!"/3#(1#*2'."-,012$ 2+2*.(0#&#+
2+2*.(0/30(2),)
!"*3(01#.04("-
*3210)/32(2
8#.2("(2)
)*,".0.#+"2&%,(0&
/3#(1#*02*0&01,*) *#(.04("! #&#+"*3,1"#*.# (2&28"2&2(4
,&."!"0/2("/(0$"1#& #1"!"*+,&"$2(1#.0+ *+,&"$('4",&%2). #&#+6).(#/,$"*011'&"1#))")/ 1#("/0++'."-'++
!"2&%,(0&"2&4!#)*2
*+,&"2:/"$2(1#.0+ $('4")#52.6
*+,&"/3#(1#*07,&2.
*'(("&2'(0%#)*"(2)
$('4) !"#&#+"#.01")/2*.(01 8#.2(")*,".2*3&0+ 2&%,(0&".0:,*0+"*321
/'-+,*"32#+.3"&'.(
2:/2(."0/,&"/3#(1#*0
-(,."!"*+,&"/3#(1#*0 !"*3(01#.04("#
#&#+"-,0#&#+"*321
#*.#"/#2$,#.(
*'(("$('4"12.#- !"*321"(2)!)
/"&'.(")0* /+#&."!
*'(("12$"(2)"0/,& *'(("12$"*321
2:/2(."0/,&",&%"$('4
*'(("/3#(1"$2),4&
*'(("$('4".#(42.) -,0.2*3&0+"-,02&4 G01!525739
-,01#))"9"-,02&2(46
K9-J=.(2B(@JFJ?,;9C.(<;0=,-.(9:(;5.($,;90:,?(!%@)(<9:D.(9F8?.F.:;,;90:(9:().8;.FH.=(STTUG((
,&/3#(1#"8227+6 )/0(.)"12$
*'((".0/"12$"*321 -,0(2)0'(*2".2*3&0+
8#.2("2&%,(0&"(2)
*'(("10+"12$ 2&%,(0&".2*3&0+ !"2:/"-0.
&'.("(2% /(0*2))"-,0*321 /+#&."*2++"#&$"2&%,(0&12&.
2:/2(."0/,&".32("/#. #//+"2&%,(0&"1,*(0- /36),0+"/+#&.#('1
#&&"#++2(4"#).31#",1 *01-"*321"3,43".")*( *'(("0(4"*321
500 #++2(46"#&$"#).31#"/(0*22$,&4)
*+,&"2:/"#++2(46
)*#&$"!"12$")*,")/0(
#++2(46
*'(("/(0.2,&"/2/.")*
!")/0(.")*,
&'.("(2)"(2% -(,."!"&'.(
80(+$"/0'+.(6")*,"!
+2.."#//+"1,*(0-,0+
!"#4("500$"*321
!"#//+"1,*(0-,0+
!"500$"/(0.2*. 500$"*321
#//+"1,*(0-,0+"-,0.
)0,+")*,")0*"#1"!
G01/5-:21;139
2&<612"#&$"1,*(0-,#+".2*3&0+046
-,0.2*3&0+"+2.. !"/+#&."/36),0+
!"#//+")/0(."/)6*30+
-(,."!")/0(."12$ .#:0& &28"/36.0+
,&."!"500$")*,".2*3 ,&."!"500$"1,*(0-,0+
0-2)")'(4
-"2&.010+"(2) D0458/1-E !")*,"500$"#4( !"500$"2&4
$0;.(;5.(?,=-.<;(F0<;(=,89>(.M8,:<90:(9:(<8,D.(J<,-.(9<(A9;5(@4G(
2(40&01,*) 500$")*,".2*3&0+!+2-
!"500$")*, )0,+"-,0+"-,0*321
&;;5739B&8/:=. 12$")*,")/0(."2:2(
-(,."/0'+.(6")*, 500$"1,*(0-,0+ !"30(.,*")*,
#1"!"*+,&"&'.( ,&"%,.(0"*2++"$2%!/+
/+#&."#&$")0,+
!"+#(6&40+"0.0+ !")0'&$"%,- *;.2/6F6!10;6!-052-5
%,/70/012 /+#&."*2++".,))"0(4
2&%,(0&"2&.010+
0 !C17/86@5+0-025!"2*0&"2&.010+
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 $2/1=1;139
EMBL-EBI storage requirements until 2008 Journal connectivity through scholarly usage data
Figure 3: Visualization of usage network created from MESUR’s 200M usage events.
(source: Andrew Lyall) (source: MESUR project)
5. USAGE-BASED METRICS on the basis of COUNTER reports, i.e. the average amount
The journal usage and citation networks also enable the of usage recorded for the articles published in a journal [4,
calculation of a variety of impact metrics. A total of 47 22].
possible impact metrics were calculated, and the resulting However, to use a social analogy, one’s importance is not
rankings were analyzed to determine the degree to which solely assessed on the basis of how many people one knows.
Cumulative GB storage usage- and citation-based metrics express similar or dissim- Who one knows and how one is embedded in a network of
ilar aspect of scholarly impact. social relationships are equally important factors. Network
theory has produced a rich literature on indicators to deter-
mine different facets of a person’s status (e.g. prestige, popu-
35000 5.1 Defining and validating usage-based met- larity, trust) on the basis of social network structure, instead
rics of using simple counts of the number of the person’s rela-
The most common indicator of journal status is Thom- tionships. Many of these indicators have found applications
son Scientific’s journal Impact Factor (IF) that is published XA
in other domains. For example, the Google search engine
30000 every year for a set of about 8,000 selected journals. The uses the PageRank metric to rank web pages on the basis
IF is defined as the average citation rate for articles pub- US
of the WWW’s hyperlink network structure. In addition,
lished in a particular journal. A similar statistical approach recent proposals have been made to rank journals according
to journal ranking has been proposed for journal usage data SR
to their citation PageRank [2] and a range of social network
25000
RF
PX
20000 OT
NM
15000 MR
MG
HC
10000
CT
DX
5000
CR
0
Se O No De Ja Fe Ma A Ma Ju J Au Se O No De Ja Fe Ma A Ma Ju J Au
p- ct- v- c- n- b- r- pr- y- n- ul- g- p- ct- v- c- n- b- r- pr- y- n- ul- g-
06 06 06 06 07 07 07 07 07 07 07 07 07 07 07 07 08 08 08 08 08 08 08 08
Cumulative storage in the National PACS since implementation in September 2006
(source: Hamish McRitchie)
!,-.(/(01(23(

6. Vol 455|4 September 2008
BOOKS & ARTS
Distilling meaning from data
Buried in vast streams of data are clues to new science. But we may need to craft new
lenses to see them, explain Felice Frankel and Rosalind Reid.
It is a breathtaking time in science they will create effective computer displays, those run by the US National Science Foun-
as masses of data pour in, prom- slides and figures for publication. Meanwhile, dation’s Picturing to Learn project (www.
ising new insights. But how can they may be developing their tools in isolation, picturingtolearn.org), teach us that attempt-
we find meaning in these tera- kept at arm’s length by scientists who are busy ing to visually communicate scientific data and
bytes? To search successfully getting their experiments done. Opportunities concepts opens a path to understanding. When
for new science in large datasets, we must find for useful dialogue are thus squandered. science and design students collaborate, their
unexpected patterns and interpret evidence When scientists, graphic artists, writers, ani- drive to understand one another’s ideas pushes
in ways that frame new questions and suggest mators and other designers come together to them to create new ways of seeing science.
further explorations. Old habits of represent- discuss problems in the visual representation Investment in visual communication training
ing data can fail to meet these challenges, pre- of science, such as at the Image and Meaning for young scientists will pay off handsomely for
venting us from reaching beyond the familiar workshops run by Harvard University (www. any data-intensive discipline.
questions and answers. imageandmeaning.org), it becomes clear The ingrained habits of highly trained sci-
To extract new meaning entists make them rarely as
D. ARMENDARIZ
from the sea of data, scien- adventurous as these young
tists have begun to embrace minds. We think we are on
23.3 Commentary Muggleton jw 20/3/06 6:29 PM Page 409
the tools of visualization. Yet the path to insight when
few appreciate that visual rep- shading reveals contours
resentation is also a form of in 3D renderings, or when
communication. A rich body bursts of red appear on heat
of communication expertise maps, for example. But the
Vol 440|23 March 2006
holds the potential to greatly algorithms used to produce
improve these tools. We pro- the graphics may create illu-
pose that graphic artists, com- sions or embed assumptions.
municators and visualization
scientists should be brought
into conversation with theo-
The human visual system
creates in the brain an appar-
ent understanding of what
COMMENTARY
rists and experimenters a picture represents, not
before all the data have been necessarily a picture of the
gathered. If we design experi- underlying science. Unless
Exceeding human limits
ments in ways that offer varied we know all the steps from
opportunities for represent- hypothesis to understand-
ing and communicating data, ing — by conversing with
techniques for extracting new theorists, experimentalists,
understanding can be made Discussing visual communication before designing experiments may reveal new science. instrument and software are turning to automated processes and technologies in a bid to cope with ever higher volumes of data.
Scientists
available. developers, visualization
But automation offers so much more to the future of science than just data handling, says Stephen H. Muggleton.
Visual representation is familiar in data- that representations repeatedly fail to com- scientists, graphic artists and cognitive psy-
intensive fields. Years before a detector is built municate understanding or address obvious chologists — we cannot be sure whether a dis-
FIREFLY PRODUCTIONS/CORBIS
for a facility such as the Large Hadron Collider questions about the underlying data. A three- play is accurate or misleading. The collection and curation
near Geneva, for example, physicists will have dimensional volume rendering may give no The greatest opportunity and risk lie in that of data throughout the
pored over simulations. They examine how hint of important uncertainties or data gaps; last step in the path: understanding. Whether sciences is becoming increas-
important events will ‘look’ in the displays solid surfaces or sharp edges may suggest data verbal or visual, any language that is garbled ingly automated. For exam-
that reveal and communicate what is going where they do not exist. A graphic artist might and inconsistent fails to do its job. Let’s talk. ple, a single high-throughput
on inside the machine. Such discussions tend propose ways to reveal gaps or deviations from Let’s all talk. I
experiment in biology can
to take place within the visual conventions of expectation early in an experiment, guiding Felice Frankel is senior research fellow in the easily generate more than a
gigabyte of data per day, and in astronomy
a field. But perhaps conversations might be subsequent data collection or highlighting new faculty of arts and sciences at Harvard University,
broadened to consider alternative represen- avenues of enquiry. When we asked Harvard Cambridge, Massachusetts 02138, USA. With data collection leads to more than a
automatic
tations of the same data. These might suggest University chemist George Whitesides to G. M. Whitesides, she is co-author of terabyte of data per night. Throughout the sci-
On the Surface
other approaches to collecting, organizing and change the geometry of a self-assembled of Things: Images of the Extraordinary in Science. volumes of archived data are increas-
ences the
querying data that will maximize the transpar- monolayer with clearly delineated hydropho- e-mail: felice_frankel@harvard.edu ing exponentially, supported not only by
ency of experimental results and thus aid intui- bic and hydrophilic areas to create an image Rosalind Reid is executive director of the Initiative storage but also by the growing
low-cost digital
tion, discovery and communication. for submission to a journal, he found himself in Innovative Computing at Harvard University of automated instrumentation. It is
efficiency
Unfortunately, visualization experts and redesigning the experiment, and unexpected and former Editor of American Scientist. that the future of science involves the
clear
communicators are often consulted only after science emerged. expansion of automation in all its aspects: data

7. c and probability cal- and charge distributionshould become easier for autonomous experimen
On such timescales it of individual molecules however, still a decade
ic provides a formal need to be integrated scientists to reproduce new experiments and becoming standard scie Vol 455|4 September 2008
gramming languages with models describ- refute their hypotheses. Despite the potentia
BOOKS & ARTS
probability calculus ing Today’s generation of microfluidic
“Owing tomachines severe danger data
the scale and rate of that incre
the interdepen- generation, computational models of
ms of probability for dency of chemical out a specific series of ume of data generation
is designed to carry
Distilling meaning from data reactions, scientific flexibility decreases in compreh
s bayesian networks.new science. But we may needHowever, but further data now require automatic
chemical
Buried in vast streams of data are clues to reactions. to craft new
stic logic’ is a formaland Rosalind Reid. be added the tool kit by developing Academic studies on the
could to this
construction and modification.”
lenses to see them, explain Felice Frankel
differences in
statements of sound mathematical under- call what one might t
It is a breathtaking time in science they will create effective computer displays, those run by the US National Science Foun-
as masses of data pour in, prom- slides and figures for publication. Meanwhile, dation’s Picturing to Learn project (www.
ising new insights. But how can they may be developing their tools in isolation, picturingtolearn.org), teach us that attempt-
a ‘chemical Turing “There is a severe danger that i
robability of A being pinnings of, say, differential equations, bayesian puter. Such chips contai
we find meaning in these tera- kept at arm’s length by scientists who are busy ing to visually communicate scientific data and
bytes? To search successfully getting their experiments done. Opportunities concepts opens a path to understanding. When
for new science in large datasets, we must find for useful dialogue are thus squandered. science and design students collaborate, their
machine’. The universal
ure forms of existing networks and logic programs make integrating chambers, ducts, gates t
unexpected patterns and interpret evidence When scientists, graphic artists, writers, ani- drive to understand one another’s ideas pushes
increases in speed and volume of n
in ways that frame new questions and suggest mators and other designers come together to them to create new ways of seeing science.
further explorations. Old habits of represent- discuss problems in the visual representation Investment in visual communication training
Turing machine, devised
fortunately computa- these various models virtually impossible. reagent stores, and allow
ing data can fail to meet these challenges, pre- of science, such as at the Image and Meaning for young scientists will pay off handsomely for
venting us from reaching beyond the familiar workshops run by Harvard University (www. any data-intensive discipline.
wever, an increasing Although by Alan Turing, be data generation could leadat high sp
in 1936 hybrid models can built by simply sis and testing to
questions and answers. imageandmeaning.org), it becomes clear The ingrained habits of highly trained sci-
t
To extract new meaning entists make them rarely as
D. ARMENDARIZ
from the sea of data, scien- adventurous as these young
tists have begun to embrace minds. We think we are on
was intended to mimic decreases in comprehensibility.”
ups have developed patching two models together, the underlying miniaturizing our robot-o
23.3 Commentary Muggleton jw 20/3/06 6:29 PM Page 409
the tools of visualization. Yet the path to insight when
few appreciate that visual rep- shading reveals contours
resentation is also a form of in 3D renderings, or when
communication. A rich body
of communication expertise
holds the potential to greatly
the pencil-and-paper
ques that can handle differences lead to unpredictable and error- this way, with the overal bursts of red appear on heat
maps, for example. But the
Vol 440|23 March 2006
algorithms used to produce
s
probabilistic logic6. prone behaviour mathematician. The chemical experimental cycle time
improve these tools. We pro-
pose that graphic artists, com- operations of a when changes are made. beings. This is particu
the graphics may create illu-
sions or embed assumptions.
municators and visualization
such research holds Turing encouraging development in this liseconds.associated with
scientists should be brought
machine would be a universal proces- nologies With microflu COMMENTARY
The human visual system
creates in the brain an appar-
One
into conversation with theo- ent understanding of what
rists and experimenters a picture represents, not
egration of scientific respect is the emergence withinbroad range of chemical reaction not onA
before all the data have been
gathered. If we design experi- sor capable of performing a computer sci- and experimentation.
necessarily a picture of the
underlying science. Unless
al and computer-sci- ence of new formalisms5 that integrate, in alimits
chemical operations Exceeding human complete, but also requi
ments in ways that offer varied we know all the steps from
opportunities for represent-
ing and communicating data,
techniques for extracting new
on both the reagents essentially human activhypothesis to understand-
ing — by conversing with
theorists, experimentalists,
available to it at the start andoffersto automated processes andof science thaninjustbid to cope with saysStephen H. Muggleton. a
thoseof mathe- of input materials, with o
Scientists are turning
chemicals bothhandling, ever higher volumes of data.
technologies a
data in the statement
understanding can be made Discussing visual communication before designing experiments may reveal new science. instrument and software
available.
sound fashion, two major branches more to the future
But automation so much developers, visualization
Visual representation is familiar in data- that representations repeatedly fail to com- scientists, graphic artists and cognitive psy-
matics: mathematical logic and probabilityauto- On such timescales it sho
it later generates. The machine would cal- clear and undeniable
intensive fields. Years before a detector is built municate understanding or address obvious chologists — we cannot be sure whether a dis-
FIREFLY PRODUCTIONS/CORBIS
for a facility such as the Large Hadron Collider questions about the underlying data. A three- play is accurate or misleading. The collection and curation
near Geneva, for example, physicists will have dimensional volume rendering may give no The greatest opportunity and risk lie in that of data throughout the
s culus. Mathematicaland test chemical com- scientists to reproduce n
matically prepare logic provides a formal experimentation.
pored over simulations. They examine how hint of important uncertainties or data gaps; last step in the path: understanding. Whether sciences is becoming increas-
important events will ‘look’ in the displays solid surfaces or sharp edges may suggest data verbal or visual, any language that is garbled ingly automated. For exam-
that reveal and communicate what is going where they do not exist. A graphic artist might and inconsistent fails to do its job. Let’s talk. ple, a single high-throughput
pounds but it would also be programmable, Stephen H. Muggleton is
learning approaches foundation for logic programming languages refute their hypotheses.
on inside the machine. Such discussions tend propose ways to reveal gaps or deviations from Let’s all talk. I
experiment in biology can
to take place within the visual conventions of expectation early in an experiment, guiding Felice Frankel is senior research fellow in the easily generate more than a
gigabyte of data per day, and in astronomy
a field. But perhaps conversations might be subsequent data collection or highlighting new faculty of arts and sciences at Harvard University,
ng scientific models such as Prolog, much theprobability calculusa Computing and the Centr
thus allowing whereas same flexibility as
broadened to consider alternative represen- avenues of enquiry. When we asked Harvard Cambridge, Massachusetts 02138, USA. With data collection leads to more than a
automatic
Today’s generation of m
tations of the same data. These might suggest University chemist George Whitesides to G. M. Whitesides, she is co-author of terabyte of data per night. Throughout the sci-
On the Surface
other approaches to collecting, organizing and change the geometry of a self-assembled of Things: Images of the Extraordinary in Science. volumes of archived data are increas-
ences the
real chemist has in the lab.
p’ systems with no provides the basic axioms of probability for is designed to carry ou Systems Biology at Imper
querying data that will maximize the transpar- monolayer with clearly delineated hydropho- e-mail: felice_frankel@harvard.edu ing exponentially, supported not only by
ency of experimental results and thus aid intui- bic and hydrophilic areas to create an image Rosalind Reid is executive director of the Initiative storage but also by the growing
tion, discovery and communication.
low-cost digital
for submission to a journal, he found himself in Innovative Computing at Harvard University of automated instrumentation. It is
efficiency
to the collection of One can think of a chemical Turing 2BZ, UK.
Unfortunately, visualization experts and redesigning the experiment, and unexpected and former Editor of American Scientist. that the future of science involves the
communicators are often consulted only after science emerged.
clear
expansion of automation in all its aspects: data

8. Released under Creative Commons License

9. Science Paradigms
T
empirical
describing natural phenomena
theoretical 2
using models, generalizations 4 2
= K
3 2
computational
simulating complex phenomena
T data exploration
unify theory, experiment, and simulation
s
FIGURE 1
CE: WHAT IS IT?
ce is where “IT meets scientists.” Researchers are using many di erent meth-
collect or generate data—from sensors and CCDs to supercomputers and
e colliders. When the data finally shows up in your computer, what do
with all this information that is now in your digital shoebox? People are
Released under Creative Commons
ually seeking me out and saying, “Help! I’ve got all this data. What am I
License

10. Principles?
CIENCE AND GOVERNMENT
POLICY FORUM
Appropriate professional and career re-
ward structures are necessary (20–22). The
An International Framework way scientists are being evaluated and how
their careers are shaped are at stake. For ex-
to Promote Access to Data ample, researchers who have spent years on
building new databases, such as the Sloan
Digital Sky Survey in astronomy, have ef-
Peter Arzberger, 1* Peter Schroeder,2 Anne Beaulieu,3 Geof Bowker,1
fectively put their scientific careers on hold
Kathleen Casey, 1 Leif Laaksonen,4 David Moorman,5 Paul Uhlir,6 Paul Wouters3
even though these databases are critical for C OV E R F E AT U RE
the future development of the
R
ecent national and multina- field. These considerations apply
tional investments (1) in OPERATING PRINCIPLES FOR DATA ACCESS REGIMES equally to those who produce,
networking and continued Openness manage, and reuse research data.
ins in information technologi- Transparency and active data dissemination At this point there is consid-
l capability (2) have given rise erable heterogeneity in policies.
Assignment and assumption of formal responsibilities
a complex cyberinfrastructure In the United States, federal
Technical and semantic interoperability of databases
at is rapidly increasing our abil- government databases are not
Downloaded from www.sciencemag.org on August 30, 2009
y to produce, manage, and use Quality control, data validation, authentication, and authorization copyright protected, whereas in
ta (3). As research becomes in- Operational efficiency and flexibility the European Union govern-
easingly global (4), data-inten- Respect for intellectual property and other ethical and legal requirements ment databases are eligible for
ve, and multifaceted (5, 6), it is Management accountability, including funding approaches protection under several data-
mperative to address national base protection laws. Even with-
d international data access and in countries, different funding
aring issues systematically in a policy are- derstanding global climate change (10) re- agencies have different stated policies; for
that transcends national jurisdictions. quires access to data drawn from many dis- example, in Canada, with three major sci-
pen access to publicly funded data pro- ciplines and sources. This issue has been a ence funding agencies, one follows the
des greater returns from the public invest- topic of recent debate and its resolution is a principles in the OECD declaration, one
ent in research, generates wealth through high priority in many scientific and policy- states access should not be a barrier, and a
ownstream commercialization of outputs, making communities (11–17). third has no policy (23). National laws and
d provides decision-makers with facts Analysis of these, and other examples international agreements can directly af-
eded to address complex, often transna- (18), suggests that successful data access fect data access and sharing practices.
THE CHANGING PARADIGM OF
onal, problems. This article summarizes and sharing arrangements exhibit a number At the last meeting of the OECD Com-
y findings of an international group that of key attributes and operating principles mittee for Scientific and Technological Poli-
udied these issues on behalf of the Organ- (see table, this page). Administrative and cy (CSTP) at the ministerial level, ministers
ation for Economic Cooperation and De- organizational management “domains” endorsed a declaration (8) based on the prin-
DATA-INTENSIVE COMPUTING
lopment (OECD) (7), which resulted in a (see figure, this page) ciple that research data
inisterial-level declaration (8). provide a framework from public funding
Legitimate restrictions on open access, for locating and ana- Technological
should be openly avail-
d strong disincentives to sharing exist, lyzing where improve- able. Furthermore, they
sed on concerns of protecting national se- ments can be made. Data access
invited OECD to devel-
Cultural Institutional
rity, privacy and confidentiality, intellec- Diversity in science and management and op a set of guidelines
al property, and time-limited exclusive use suggests that a variety behaviorial domains managerial based on commonly
the scientific investigator. The lack of of institutional models agreed principles (simi-
ear funding-agency policies in the face of and tailored data man- L
Legal Financial
an lar to those in the table)
and and
rong competing interests, often far re- agement approaches policy budgetary to facilitate optimal
oved from academic research, poses prob- will be needed. cost-effective access to
ms for scientists in developing and devel- Establishing and Domains of a data access regime. digital research data Richard T. Kouzes, Gordon A. Anderson, Stephen T. Elbert, Ian Gorton, and
ped countries and inhibit the advance of maintaining this infra- from public funding. It Deborah K. Gracio, Pacific Northwest National Laboratory
ience for the public good. For example, structure requires continued and dedicated can be expected that these future guidelines
search on cholera outbreaks and their rela- budgetary planning, with appropriate fi- will influence national and international reg-
on to environmental factors (9) or on un- nancial support. The use of research data ulation of research data, much as the OECD
cannot be maximized if access, manage- Guidelines on the Protection of Privacy (24),
University of California, San Diego, La Jolla, CA 92093, ment, and preservation costs (including which have been a model for legislation all Through the development of new classes of erogeneous full-scale simulations will require not only
SA. 2Ministry of Education, Culture and Science,
cost of documentation and metadata cre- around the Western world. peta op capabilities but also a computational infrastruc-
oetermeer, Netherlands. 3Networked Research and ation) are an afterthought or are insuffi- Although the involvement of re-
software, algorithms, and hardware, data-
gital Information, Royal Netherlands Academy of Arts
ciently or inconsistently funded in research searchers in resolving these issues is criti- intensive applications provide timely and ture that permits model integration. Simultaneously, it
nd Sciences, Amsterdam, Netherlands. 4CSC-Scientific
projects (19). D. Atkins et al. (3) recom- cal, many scientists remain ignorant about must couple to huge databases created by an ever-in-
omputing Ltd., Espoo, Finland. 5Social Sciences and Hu-
anities Research Council, Ottawa, Canada. 6National mend that roughly one-third of the provi- existing policies at their institutions or na- meaningful analytical results in response
creasing number of high-throughput instruments.”2
search Council, Washington, DC 20418, USA. sioning and operations of cyberinfrastruc- tions, let alone those of other countries. To to exponentially growing data complexity

11. An International Framework way scientists are being
their careers are shaped
to Promote Access to Data ample, researchers who
Principles?
building new databases
Digital Sky Survey in
Peter Arzberger, 1* Peter Schroeder,2 Anne Beaulieu,3 Geof Bowker,1
fectively put their scien
Kathleen Casey, 1 Leif Laaksonen,4 David Moorman,5 Paul Uhlir,6 Paul Wouters3
even though these datab
the future d
R
ecent national and multina- field. These c
tional investments (1) in OPERATING PRINCIPLES FOR DATA ACCESS REGIMES equally to t
networking and continued Openness manage, and
gains in information technologi- Transparency and active data dissemination At this po
cal capability (2) have given rise
to a complex cyberinfrastructure
CIENCE AND GOVERNMENT
POLICY FORUM
Assignment and assumption of formal responsibilities
Technical and semantic interoperability of databases
erable hetero
In the Unit
that is rapidly increasing our abil- Appropriate professional and career re-
ward structures are necessary (20–22). The
government
An International Framework
ity to produce, manage, and use Quality control, data validation, authentication, and authorization
way scientists are being evaluated and how
their careers are shaped are at stake. For ex- copyright pr
to Promote Access to Data research becomes in-
data (3). As Operational efficiency and flexibility
ample, researchers who have spent years on
building new databases, such as the Sloan
Digital Sky Survey in astronomy, have ef-
the Europe
1* Peter Schroeder,2 Anne Beaulieu,3 Geof Bowker,1
creasingly global (4), data-inten- ment databa
Peter Arzberger, fectively put their scientific careers on hold
Kathleen Casey, Respect F E ATintellectual property and other ethical and legal requirements
C OV E R for U RE
1 Leif Laaksonen,4 David Moorman,5 Paul Uhlir,6 Paul Wouters3
even though these databases are critical for
the future development of the
sive, and multifaceted (5, 6), it is protection u
R
ecent national and multina- field. These considerations apply
tional investments (1) in Management accountability, including funding approaches
OPERATING PRINCIPLES FOR DATA ACCESS REGIMES equally to those who produce,
imperative to address national base protecti
networking and continued Openness manage, and reuse research data.
ins in information technologi- Transparency and active data dissemination At this point there is consid-
l capability (2) have given rise erable heterogeneity in policies.
and international data access and in countries
Assignment and assumption of formal responsibilities
a complex cyberinfrastructure In the United States, federal
Technical and semantic interoperability of databases
at is rapidly increasing our abil- government databases are not
Downloaded from www.sciencemag.org on August 30, 2009
y to produce, manage, and use Quality control, data validation, authentication, and authorization copyright protected, whereas in
sharing issues systematically in a policy are- derstanding global climate change (10) re- agencies have different
ta (3). As research becomes in-
easingly global (4), data-inten-
Operational efficiency and flexibility
Respect for intellectual property and other ethical and legal requirements
the European Union govern-
ment databases are eligible for
na that transcends national jurisdictions. quires access to data drawn from many dis- example, in Canada, w
ve, and multifaceted (5, 6), it is
mperative to address national
Management accountability, including funding approaches protection under several data-
base protection laws. Even with-
d international data access and in countries, different funding
Open access to publicly funded data pro- ciplines and sources. This issue has been a ence funding agencies
aring issues systematically in a policy are- derstanding global climate change (10) re- agencies have different stated policies; for
that transcends national jurisdictions. quires access to data drawn from many dis- example, in Canada, with three major sci-
vides greater returns from the public invest- topic of recent debate and its resolution is a principles in the OEC
pen access to publicly funded data pro- ciplines and sources. This issue has been a ence funding agencies, one follows the
des greater returns from the public invest- topic of recent debate and its resolution is a principles in the OECD declaration, one
ent in research, generates wealth through high priority in many scientific and policy- states access should not be a barrier, and a
ment in research, generates wealth through high priority in many scientific and policy- states access should no
ownstream commercialization of outputs, making communities (11–17).
d provides decision-makers with facts
third has no policy (23). National laws and
Analysis of these, and other examples international agreements can directly af-
downstream commercialization of outputs, making communities (11–17).
eded to address complex, often transna- (18), suggests that successful data access fect data access and sharing practices.
third has no policy (23)
THE CHANGING PARADIGM OF
onal, problems. This article summarizes and sharing arrangements exhibit a number At the last meeting of the OECD Com-
y findings of an international group that of key attributes and operating principles mittee for Scientific and Technological Poli-
and provides decision-makers with facts Analysis of these, and other examples international agreemen
udied these issues on behalf of the Organ- (see table, this page). Administrative and cy (CSTP) at the ministerial level, ministers
ation for Economic Cooperation and De- organizational management “domains” endorsed a declaration (8) based on the prin-
needed to address complex, often transna- (18), suggests that successful data access fect data access and sha
DATA-INTENSIVE COMPUTING
lopment (OECD) (7), which resulted in a (see figure, this page) ciple that research data
inisterial-level declaration (8). provide a framework from public funding
Legitimate restrictions on open access, for locating and ana- Technological
should be openly avail-
tional, problems. This article summarizes and sharing arrangements exhibit a number
d strong disincentives to sharing exist, lyzing where improve-
sed on concerns of protecting national se- ments can be made.
Cultural Data access Institutional
invited OECD to devel-At the last meeting
able. Furthermore, they
rity, privacy and confidentiality, intellec- Diversity in science op a set of guidelines
key findings of an international group that of key attributes and operating principles mittee for Scientific and
al property, and time-limited exclusive use suggests that a variety
the scientific investigator. The lack of of institutional models
and
behaviorial
management
domains
and
managerial based on commonly
agreed principles (simi-
studied these issues on behalf of the Organ- (see table, this page). Administrative and cy (CSTP) at the minist
ear funding-agency policies in the face of and tailored data man-
rong competing interests, often far re- agement approaches
oved from academic research, poses prob- will be needed.
L
Legal
and
policy
Financial
an
and
budgetary
lar to those in the table)
to facilitate optimal
cost-effective access to
isation for Economic Cooperation and Richard T. Kouzes, GordonNorthwest National Laboratory Ian Gorton,“domains” endorsed a declaration (
ms for scientists in developing and devel-
ped countries and inhibit the advance of maintaining this infra-
De- organizational Stephen T. Elbert,
Deborah K. Gracio, Pacific
A. Anderson,
management and
Establishing and Domains of a data access regime. digital research data
from public funding. It
velopment (OECD) (7), which resulted in a (see figure, this page)
ience for the public good. For example, structure requires continued and dedicated can be expected that these future guidelines
search on cholera outbreaks and their rela- budgetary planning, with appropriate fi- will influence national and international reg-
on to environmental factors (9) or on un- nancial support. The use of research data ulation of research data, much as the OECD
cipl
ministerial-level declaration (8). Through the development of a framework
provide new classes of erogeneous full-scale simulations will require not only from
cannot be maximized if access, manage- Guidelines on the Protection of Privacy (24),
University of California, San Diego, La Jolla, CA 92093, ment, and preservation costs (including which have been a model for legislation all
Legitimate restrictions on open access, applications and hardware, data- ana- that permits model integration. Simultaneously, it
for locating and
cost of documentation and metadata cre- around the Western world. Technological
sho
SA. 2Ministry of Education, Culture and Science,
oetermeer, Netherlands. 3Networked Research and software, algorithms, peta op capabilities but also a computational infrastruc-
ation) are an afterthought or are insuffi- Although the involvement of re-
gital Information, Royal Netherlands Academy of Arts
intensive provide timely and ture
ciently or inconsistently funded in research searchers in resolving these issues is criti-
meaningful lyzing where improve-
and strong disincentives to sharing exist, analytical results in response must couple to huge databases created by an ever-in- able
nd Sciences, Amsterdam, Netherlands. 4CSC-Scientific
omputing Ltd., Espoo, Finland. 5Social Sciences and Hu- projects (19). D. Atkins et al. (3) recom- cal, many scientists remain ignorant about
anities Research Council, Ottawa, Canada. 6National mend that roughly one-third of the provi- existing policies at their institutions or na-
creasing number of high-throughput instruments.” 2
search Council, Washington, DC 20418, USA. sioning and operations of cyberinfrastruc- tions, let alone those of other countries. To to exponentially growing data complexity

12. An International Framework
• Approximations are currently made because of limitations on spatial scale; more accurate
way scientists are being
their careers are shaped
algorithms are needed.
to Promote Access to Data ample, researchers who
Principles?
building new databases
• Need a universal parser to tag information for analysis. Digital Sky Survey in
Peter Arzberger, 1* Peter Schroeder,2 Anne Beaulieu,3 Geof Bowker,1
fectively put their scien
Kathleen Casey, 1 Leif Laaksonen,4 David Moorman,5 Paul Uhlir,6 Paul combining all available data.
• Need algorithms to recognize and predict intent by Wouters3 even though these datab
Information analytics • Require an integrating informatics resource manager that takes in sensor data, transforming d the future
R
ecent national andbetween heterogeneous datasets and integrating computational tools, and presents results to c
multina- field. These
tional investments (1) in OPERATING PRINCIPLES FOR DATA ACCESS REGIMES equally to t
the human user.
networking and continued Openness manage, and
gains in information technologi- scale of data and problems overwhelms visual displays, new approaches are this po
• Since the Transparency and active data dissemination At
CIENCE AND GOVERNMENT
POLICY FORUM
cal capability (2) have given rise develop the appropriate level of abstraction and to condense and select data to dis-
needed to Assignment and assumption of formal responsibilities
to a complex cyberinfrastructure the user’s control.
play under Technical and semantic interoperability of databases
erable hetero
In the Unit
that is rapidly increasing our abil- Appropriate professional and career re-
ward structures are necessary (20–22). The
government
An International Framework
ity to produce, manage,• CollaborativeQuality control,analysis of datasets and observations are desirable.
and use data validation, authentication, and authorization
way scientists are being evaluated and how
copyright pr
sharing and their careers are shaped are at stake. For ex-
to Promote Access to Data research becomes in-
data (3). As Operational efficiency and flexibility
ample, researchers who have spent years on
building new databases, such as the Sloan
Digital Sky Survey in astronomy, have ef-
the Europe
1* Peter Schroeder,2 Anne Beaulieu,3 Geof Bowker,1
Respect F ATintellectual propertythe problem set to access large local andment databa
creasingly global (4), data-inten- architecturesEareRE
Peter Arzberger,
Computing platforms • Current C OV E R for U inadequate for and other ethical and legal requirements distributed
fectively put their scientific careers on hold
Kathleen Casey, 1 Leif Laaksonen,4 David Moorman,5 Paul Uhlir,6 Paul Wouters3
even though these databases are critical for
the future development of the
sive, and multifaceted (5, 6), it is providing solutions with reasonable throughput to analyze and model atprotection u
datasets, the
R
ecent national and multina- field. These considerations apply
tional investments (1) in Management accountability, including funding approaches
OPERATING PRINCIPLES FOR DATA ACCESS REGIMES equally to those who produce,
imperative to address required spatial scales.
national base protecti
networking and continued Openness manage, and reuse research data.
ins in information technologi- Transparency and active data dissemination At this point there is consid-
l capability (2) have given rise erable heterogeneity in policies.
and international data access and in countries
Assignment and assumption of formal responsibilities
a complex cyberinfrastructure In the United States, federal
Technical and semantic interoperability of databases
at is rapidly increasing our abil- government databases are not
Downloaded from www.sciencemag.org on August 30, 2009
sharing issues systematically in a policy are- not have the storage climate change amount of data that models use
• Current machines do derstanding global capacity for the (10) re-
y to produce, manage, and use Quality control, data validation, authentication, and authorization copyright protected, whereas in
ta (3). As research becomes in-
easingly global (4), data-inten-
Operational efficiency and flexibility
Respect for intellectual property and other ethical and legal requirements
the European Union govern-
ment databases are eligible for
agencies have different
na that transcends national jurisdictions. quires access to data drawn from many dis- example, in Canada, w
ve, and multifaceted (5, 6), it is
mperative to address national and produce.
Management accountability, including funding approaches protection under several data-
base protection laws. Even with-
d international data access and in countries, different funding
Open access to publicly funded data pro- ciplines and sources. This issue has been a ence funding agencies
aring issues systematically in a policy are- derstanding global climate change (10) re- agencies have different stated policies; for
that transcends national jurisdictions. quires access to data drawn from many dis- example, in Canada, with three major sci-
vides greater returns from the public invest- topic of recent debate and its systems with high-performance net- OEC
• Need self-healing and intrinsically secure operating resolution is a principles in the
pen access to publicly funded data pro- ciplines and sources. This issue has been a ence funding agencies, one follows the
des greater returns from the public invest- topic of recent debate and its resolution is a principles in the OECD declaration, one
ment in research, generates wealth through built-in encryption.
working that provides high priority in many scientific and policy- states access should no
ent in research, generates wealth through high priority in many scientific and policy- states access should not be a barrier, and a
ownstream commercialization of outputs, making communities (11–17). third has no policy (23). National laws and
d provides decision-makers with facts Analysis of these, and other examples international agreements can directly af-
downstream commercialization of outputs, making communities (11–17).
eded to address complex, often transna- (18), suggests that successful data access fect data access and sharing practices.
third has no policy (23)
THE CHANGING PARADIGM OF
onal, problems. This article summarizes and sharing arrangements exhibit a number At the last meeting of the OECD Com-
• Computational needs range from largethese, and other examples
and provides decision-makers with facts Analysis of central high-performance computing systems to
y findings of an international group that of key attributes and operating principles mittee for Scientific and Technological Poli-
udied these issues on behalf of the Organ- (see table, this page). Administrative and cy (CSTP) at the ministerial level, ministers
international agreemen
ation for Economic Cooperation and De- organizational management “domains” endorsed a declaration (8) based on the prin-
needed to address complex, often transna- systemssuggestsminiaturized labs on a chip for eld deployments. and sha
portable lightweight (18), such as that successful data access fect data access
DATA-INTENSIVE COMPUTING
lopment (OECD) (7), which resulted in a (see figure, this page) ciple that research data
inisterial-level declaration (8). provide a framework from public funding
Legitimate restrictions on open access, for locating and ana- Technological
should be openly avail-
tional, problems. This article summarizes and sharing arrangements exhibit a number
d strong disincentives to sharing exist, lyzing where improve-
sed on concerns of protecting national se- ments can be made.
Cultural Data access Institutional
able. Furthermore, they
invited OECD to devel- At the last meeting
rity, privacy and confidentiality, intellec- Diversity in science op a set of guidelines
key findings of an international group that of key attributes and operating principles mittee for Scientific and
al property, and time-limited exclusive use suggests that a variety
the scientific investigator. The lack of of institutional models
and
behaviorial
management
domains
and
managerial based on commonly
agreed principles (simi-
studied these issues on behalf of the Organ- (see table, this page). Administrative and cy (CSTP) at the minist
ear funding-agency policies in the face of and tailored data man-
rong competing interests, often far re- agement approaches
oved from academic research, poses prob- will be needed.
L
Legal
and
policy
Financial
an
and
budgetary
lar to those in the table)
to facilitate optimal
cost-effective access to
isation for Economic Cooperation and Richard T. Kouzes, GordonNorthwest National Laboratory Ian Gorton,“domains” endorsed a declaration (
ms for scientists in developing and devel-
ped countries and inhibit the advance of maintaining this infra-
De- organizational Stephen T. Elbert,
Deborah K. Gracio, Pacific
A. Anderson,
management and
Establishing and Domains of a data access regime. digital research data
from public funding. It
COMPUTER velopment (OECD) (7), which resulted in a (see figure, this page)
ience for the public good. For example, structure requires continued and dedicated can be expected that these future guidelines
search on cholera outbreaks and their rela- budgetary planning, with appropriate fi- will influence national and international reg- cipl
on to environmental factors (9) or on un- nancial support. The use of research data ulation of research data, much as the OECD
ministerial-level declaration (8). Through the development of a framework
provide new classes of erogeneous full-scale simulations will require not only from
cannot be maximized if access, manage- Guidelines on the Protection of Privacy (24),
University of California, San Diego, La Jolla, CA 92093, ment, and preservation costs (including which have been a model for legislation all
Legitimate restrictions on open access, applications and hardware, data- ana- that permits model integration. Simultaneously, it
for locating and
cost of documentation and metadata cre- around the Western world. Technological
sho
SA. 2Ministry of Education, Culture and Science,
oetermeer, Netherlands. 3Networked Research and software, algorithms, peta op capabilities but also a computational infrastruc-
ation) are an afterthought or are insuffi- Although the involvement of re-
gital Information, Royal Netherlands Academy of Arts
intensive provide timely and ture
ciently or inconsistently funded in research searchers in resolving these issues is criti-
meaningful lyzing where improve-
and strong disincentives to sharing exist, analytical results in response must couple to huge databases created by an ever-in- able
nd Sciences, Amsterdam, Netherlands. 4CSC-Scientific
omputing Ltd., Espoo, Finland. 5Social Sciences and Hu- projects (19). D. Atkins et al. (3) recom- cal, many scientists remain ignorant about
anities Research Council, Ottawa, Canada. 6National mend that roughly one-third of the provi- existing policies at their institutions or na-
creasing number of high-throughput instruments.” 2
search Council, Washington, DC 20418, USA. sioning and operations of cyberinfrastruc- tions, let alone those of other countries. To to exponentially growing data complexity

13. Computional Domain
Thinkers Specialists
Creating
Formulation Interaction
Data models & Experiments &
computational knowledge
methods creation
Mapping Steering
Data-Intensive
Engineers
Execution
Implementations,
compute & data
resources

14. Efficient distributed
systems
Computer Science
Research
Effective
algorithms Data-intensive
computing

15. Efficient distributed Reusable computational
systems models
Computer Science Interdisciplinary
Research Applications
Effective Intuitive
algorithms Data-intensive Collaborative interfaces
computing environments
New conceptual
models for systems

16. Developmental Medical Emergency
Chemistry Response
Biology Genetics
Reusable computational
models
alpha release of a combined earth-
quake selection and waveform selec-
tion service combining the EMSC and
Real-time access to European BB
data successively increasing
The Virtual European Broad-band
the ORFEUS services. The web por- Seismograph Network (VEBSN) is
tal also includes a first test version steadily increasing its size. Currently
of the underlying software structure more then 270 stations are contrib-
Interdisciplinary
of the distributed archive services of uting data to the VEBSN in near real-
the Integrated European Distributed time. For some tens of these stations
Archive (EIDA) for waveform data. we still need to compile the instru-
The alpha release implies that a mentation and data details (data-
test version of the current service is less Seed volumes). An example of
made accessible for a selected group the earthquake in Greece on Febru-
Applications
of scientist that are willing to test it ary 14, 2008 illustrates the available
and recommend modifications. In- data. The VEBSN is a joint initiative
terested seismologists, student, re- of European-Mediterranean seismo-
searcher or network operator, are logical networks. More information
encouraged to contact the NERIES can be obtained from www.orfeus-
Project Office if they are interested eu.org/Data-info/vebsn.html.
to test the services. A short video
Intuitive
presentation is available (http:// Figure 3. The Greek earthquake of February 14, 2008
as recorded by the vertical component of broadband
www.neries-eu.org/main.php/demo. stations of the VEBSN (mainly in the European-Medi-
terranean area) and made available by ORFEUS. The
wmv?fileitem=8798210). Alessan- VEBSN is currently still expanding.
Collaborative Brain
dro Spinuso, Sergio Rives, Luca Tra-
Neuro- Quantitative
ni, Phetaphone Thomy, Rémy Bossu,
interfaces
Seismology
Torild van Eck. (See figure 2 below.)
informatics Genetics Imaging
environments

17. Computional Domain
Thinkers Specialists
Creating
Formulation Interaction
Data models & Experiments &
computational knowledge
methods creation
Mapping Steering
Data-Intensive
Engineers
Execution
Implementations,
compute & data
resources

18. Interaction
Experiments &
knowledge
creation
Steering
Data-Intensive
Engineers
Execution

19. Interaction
Experiments &
knowledge
creation
Steering
Data-Intensive
Engineers
Execution

20. XML
interface,
task &
resource
description
1. specifies 2. uses
students teacher
3. deploys
portal 4. performs task researcher
designer
9. analyses web portal
results
5. configures
6. runs jobs
7. monitors
8. returns
results
!"#$%&"'(
compute resources

22. Computional Domain
Thinkers Specialists
Creating
Formulation Interaction
Data models & Experiments &
computational knowledge
methods creation
Mapping Steering
Data-Intensive
Engineers
Execution
Implementations,
compute & data
resources

23. Formulation
Data models &
computational
methods
Mapping
Data-Intensive
Engineers
Execution

24. Formulation
Data models &
computational
methods
Classification of Gene
MappingPatterns
Expression
Data-Intensive
Engineers
Execution

25. Testing phase Training phase
Manual Image Image
Annotations integration processing
Image
processing
Feature
Feature generation
Images generation
Feature Feature
selection/ selection/
Deployment phase extraction
extraction
Apply
classifier
Automatic Prediction Classifier
annotations evaluation construction

26. can be continuously and the initial input signal therefore is decomposed into
different subbands.
LL2 LH2
LH1
Wavelet
decomposition
HL2 HH2
2 LL2out
LL
2 HL2out
HL
2 LL1out
LL HL1 HH1
2 LH2out
HL1out LH
2
HL
2D-Array 2 HH2out
HH
Input 2 LH1out
LH
2 HH1out
HH
(a) Wavelet decomposition on 2D-array (b) Wavelet decomposition on an image
Fig. 2. Wavelet decomposition
Liangxiu Han
Mathematically, for a signal f (x, y) with 2D array(M ∗ N ), the wavele

27. User and application diversity
Iterative DMI
process
development Accommodating
Tool level
Many application domains
Many tool sets
Many process representations
Many working practices
Gateway interface
DMI canonical representation and abstract machine
one model
Composing or hiding
Enactment Many autonomous resources & services
level Mapping
optimisation Multiple enactment mechanisms
and Multiple platform implementations
enactment
System diversity and complexity

28. User and application diversity
Iterative DMI
process
development Accommodating
Tool level
Many application domains
Many tool sets
Many process representations
Many working practices
Gateway interface
DMI canonical representation and abstract machine
one model
Composing or hiding
Enactment Many autonomous resources & services
level Mapping
optimisation Multiple enactment mechanisms
and Multiple platform implementations
enactment
System diversity and complexity

29. Testing phase Training phase
Manual Image Image
Annotations integration processing
Image
processing
Feature
Formulation
Feature generation
Images generation
OGSA-DAI
Feature Feature
selection/ selection/
Deployment phase extraction
extraction
Apply
classifier
Automatic Prediction Classifier
annotations evaluation construction
Data-Intensive Systems Process
/* import non-universal components from the computational environment */
import uk.org.ogsadai.SQLQuery; //get definition of SQLQuery
Engineering Language
import uk.org.ogsadai.TupleToWebRowSetCharArrays; // serialisation
import uk.org.ogsadai.DeliverToRequestStatus;
/* construct and identify instances of the PE */
SQLQuery query = new SQLQuery();
Java
TupleToWebRowSetCharArrays wrs = new TupleToWebRowSetCharArrays();
DeliverToRequestStatus del = new DeliverToRequestStatus();
/* form connection c1 with an explicit literal stream expression as its source
and query as its destination */
String q1 = "SELECT * FROM weather";
|- q1 -| => expression->query;
String resourceID = "MySQLResource";
|- resourceID -| => resource->query;
query->data => data->wrs;
wrs->result => input->del;

30.  













     


  
    
  
  
  
  

  








31. 
 


 


 

 




 
     




  
      
  
  
  
   

   



     





  


  
  
     
    
    
  

 
 
  
 
   



   

   

  
 
   
     
 


      
    
   
   
 



 
      

   
   
   
       

 
 
    
       
  
  
      
 

     

   
    
 
 
 
 

   
   
   
  
 
 

       
  

 




    



32. Architecture results
5
2 nodes
3 nodes
4.5 4 nodes
5 nodes
6 nodes
4 7 nodes
8 nodes
3.5
Speedup
3
2.5
2
1.5
1
0.5
0 5000 10000 15000 20000
Number of Images

33. The ‘hump’
6000
Workflow Execution Time
PE1
PE2
5000 PE3
PE4
PE5
Processing Time(s)
PE6
4000 PE7
PE8
PE9
3000
2000
1000
0
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
6400 Images 12800 Images 19200 Images
Number of Computing Nodes

34. The ‘hump’
6000
Workflow Execution Time
PE1
PE2
5000 PE3
PE4
PE5
Processing Time(s)
PE6
4000 PE7
PE8
PE9
3000
2000
1000
0
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
6400 Images 12800 Images 19200 Images
Number of Computing Nodes

35. No ‘hump’
5000 Workflow Execution Time
PE1
PE2
PE3
PE4
Workflow execution time
4000 PE5
Processing Time(s)
PE6
PE7
PE8
PE9
3000
2000
1000
0
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
6400 Images 12800 Images 19200 Images
Number of Computing Nodes

36. No ‘hump’
5000 Workflow Execution Time
PE1
PE2
PE3
PE4
Workflow execution time
4000 PE5
Processing Time(s)
PE6
PE7
PE8
PE9
3000
2000
1000
0
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
6400 Images 12800 Images 19200 Images
Number of Computing Nodes

37. Data mining results
Table 1. The preliminary result of classification performance using 10-fold validation
hhhh
h hhClassification Performance
hhhh
hhhh Sensitivity Specificity
Gene expression hh h
Humerus 0.7525 0.7921
Handplate 0.7105 0.7231
Fibula 0.7273 0.718
Tibia 0.7467 0.7451
Femur 0.7241 0.7345
Ribs 0.5614 0.7538
Petrous part 0.7903 0.7538
Scapula 0.7882 0.7099
Head mesenchyme 0.7857 0.5507
Note: Sensitivity: true positive rate. Specificity: true negative rate.
5 Conclusion and Future Work

38. Computional Domain
Thinkers Specialists
Creating
Formulation Interaction
Data models & Experiments &
computational knowledge
methods creation
Mapping Steering
Data-Intensive
Engineers
Execution
Implementations,
compute & data
resources

39. D
Sp
Creating
n Inte
& Exp
l kn
c

40. Spatial atlases for
developmental biology
D
Sp
Creating
n Inte
& Exp
l kn
c

41. Next Generation
Embryology
Google Maps for
Developmental Biology

42. Next generation technology
Current Repository Enhanced Repository
Query,
Navigation &
Contribution
Query &
Contribution Navigation
Web & Java clients
e-MouseAtlas e-MouseAtlas+
Web clients
DSpace + webviewer
Knowledge + Maps Knowledge Maps
Gagarine Yaikhom

44. Annotating on-line

45. Data-intensive
research workshop
Monday, 15 March 2010 @ e-Science Edinburgh
Volume Complexity Interaction
Databases
Pri
nci
Paradigms ple
Analysis
s?

46. Data-intensive Research Group
Academics
Malcolm Atkinson
Research Assistants
Jos Koetsier
Liangxiu Han
David Rodriguez
Gagarine Yaikhom
PhD Students
Thomas French
Luna De Ferrari
Rob Kitchen
Chee-Sun Liew
Fan Zhu
Research Students
Gary, Vijay, Hwee, Yue,
Charalampos, http://research.nesc.ac.uk/
Gideon, Jeff, Gareth,
Charis, Andrejs Jano van Hemert