SourceData: Making Data Discoverable

Nancy George *, Robin Liechti *, Sara El-Gebali *, Lou Götz *, Isaac Crespo ,
                     2,                     1,                       2,                1,                      1



Ioannis Xenarios , Thomas Lemberger
                          1                            1, 3




*contributed equally,
1
 Vital-IT, Swiss Institute of Bioninformatics, Lausanne Switzerland
2
 SourceData, EMBO, Heidelberg, Germany
3
 Correspondence to: thomas.lemberger@embo.org)

In molecular and cell biology, most of the data presented in published papers are not
available in formats that allow for direct analysis and systematic mining. The goal of
the SourceData project (http://sourcedata.embo.org) is to make published data easier
to find, to connect papers containing related information and to promote the reuse and
novel analysis of published data. The main concept underlying the project is that the
structure of a dataset provides information about the design of the study in question
and can be exploited in powerful data-oriented search strategies. SourceData has
therefore developed tools to generate machine-readable descriptive metadata from
figures in published manuscripts. Experimentally tested hypotheses are represented as
directed relationships between standardized biological entities. Once processed, a	
  
comprehensive	
  ‘scientific	
  knowledge	
  graph’	
  can	
  be	
  generated	
  from	
  this	
  data	
  (see	
  
demo	
  video1	
  at	
  https://vimeo.com/sourcedata/kg),	
  making	
  the	
  body	
  of	
  data	
  
efficiently	
  searchable.	
  Importantly,	
  this	
  graph	
  is	
  objectively	
  grounded	
  in	
  
published	
  data	
  and	
  not	
  on	
  the	
  potentially	
  subjective	
  interpretation	
  of	
  the	
  results.	
  	
  
	
  
SourceData	
  has	
  developed	
  algorithms	
  to	
  efficiently	
  search	
  the	
  data-­‐oriented	
  
knowledge	
  graph	
  and	
  an	
  interface,	
  shown	
  in	
  Figure	
  1,	
  that	
  enables	
  users	
  to	
  find	
  
paper	
  based	
  on	
  their	
  data	
  content:	
  

Figure 1. SourceData Search interface




The search capabilities have also been incorporated into the SmartFigure viewer. This
application can	
  be	
  embedded	
  directly	
  into	
  online	
  publications	
  and	
  allows	
  the	
  
visualisation	
  of	
  figures	
  in	
  the	
  context	
  of	
  related	
  data	
  published	
  elsewhere.	
  
Readers	
  can	
  then	
  navigate	
  from	
  one	
  figure	
  to	
  the	
  next	
  by	
  following	
  linked	
  
entities	
  (see	
  Figure	
  2	
  and	
  see	
  demo	
  video2	
  at	
  
http://vimeo.com/sourcedata/search).	
  	
  
	
  
Figure	
  2.	
  ‘SmartFigures”	
  viewer	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
	
  
Access	
  to	
  the	
  SourceData	
  database	
  by	
  computer	
  programs	
  is	
  provided	
  through	
  a	
  
public	
  Application	
  Program	
  Interface	
  (http://sourcedata.vital-­‐
it.ch/public/#/api),	
  giving	
  developers	
  the	
  chance	
  to	
  produce	
  their	
  own	
  software	
  
solutions	
  or	
  machine-­‐driven	
  analyses	
  based	
  on	
  the	
  SourceData	
  data	
  format.	
  
	
  
Future	
  perspectives	
  for	
  the	
  project	
  include	
  the	
  integration	
  of	
  a	
  structured	
  
representation	
  of	
  time	
  and	
  incorporating	
  descriptions	
  of	
  experimental	
  
procedures	
  and	
  reagents.	
  Furthermore,	
  SourceData	
  will	
  develop	
  portable	
  
‘figure/data	
  packages’	
  that	
  combine	
  and	
  cross-­‐link	
  the	
  human-­‐interpretable	
  
figure	
  to	
  the	
  underlying	
  machine-­‐readable	
  metadata	
  and	
  data	
  files.	
  	
  This	
  means	
  
linking	
  the	
  original	
  experimental	
  dataset	
  with	
  the	
  representative	
  experimental	
  
figure	
  to	
  allow	
  ease	
  of	
  re-­‐use	
  and	
  transparency	
  of	
  data.	
  	
  Finally,	
  we	
  plan	
  to	
  	
  adapt	
  
the	
  SourceData	
  model	
  to	
  integrate	
  existing	
  approaches	
  for	
  the	
  representation	
  of	
  
large-­‐scale	
  biological	
  data.	
  
	
  
With	
  SourceData,	
  we	
  are	
  developing	
  a	
  platform	
  that	
  simultaneously	
  improves	
  
the	
  discoverability	
  and	
  utility	
  of	
  research	
  data	
  and	
  of	
  the	
  scientific	
  articles	
  
where	
  these	
  data	
  are	
  reported.	
  It	
  will	
  therefore	
  provide	
  the	
  basis	
  for	
  a	
  reward	
  
system	
  that	
  will	
  incentivize	
  authors	
  to	
  share	
  their	
  data	
  openly,	
  thus	
  driving	
  a	
  
broader	
  adoption	
  of	
  open	
  data	
  and	
  open	
  science	
  by	
  the	
  community.