Here we describe simple heuristics for extracting the text pertaining to a specific experiment from a pathway papers.
Typically, experimental papers are split into sections. This is evident from the structure of the XML of the full text articles. A typical structure would be:
As an example, let’s take http://www.ncbi.nlm.nih.gov/pubmed/23142775, which is one of the Open Access papers from PL. This is interesting because in Figure 5a and 5b, the paper talks about data from a ‘coprecipitation’ assay, which is a technique used to examine what proteins complexes are made of.
This is the assay we’re going to look at in detail for the documents in the PL corpus. The XML of the results section of the PubMedCentral articles reads thus:
<sec sec-type="results" id="S1">
<title>RESULTS</title>
...
<sec id="S5">
<title>DDX41-dependent signaling downstream of c-di-GMP and c-di-AMP</title>
<p id="P10">
DDX41 interacts and co-localizes with the STING adaptor (<bold><xref
ref-type="supplementary-material" rid="SD1">Supplementary Fig. 4a,b</xref></bold>)
to facilitate DNA ligand dependent signal transduction<sup><xref ref-type="bibr" rid="R17">17</xref></sup>.
Introduction of either c-di-GMP or c-di-AMP into D2SC cells led to enhanced DDX41-STING complex
formation (<bold><xref ref-type="fig" rid="F5">Fig. 5a</xref></bold>). In DNA dependent signaling
pathways, STING further binds to the downstream kinase TBK1 to activate the type I IFN
response<sup><xref ref-type="bibr" rid="R18">18</xref>, <xref ref-type="bibr" rid="R19">19</xref></sup>.
Both c-di-GMP and c-di-AMP when transfected into control D2SC cells activated the formation of a STING-
TBK1 complex, however, c-di-GMP and c-di-AMP mediated activation of the STING-TBK1 complex was
almost completely abrogated in DDX41-shRNA cells (<bold><xref ref-type="fig" rid="F5">Fig. 5b</xref>
</bold>). Consequently, c-di-GMP and c-di-AMP mediated activation of TBK1, IRF3 and the downstream type
I IFN effector, STAT1 was impaired in DDX41-shRNA cells (<bold><xref ref-type="fig" rid="F5">Fig. 5c</xref>
and <xref ref-type="supplementary-material" rid="SD1">Supplementary Fig. 5</xref></bold>). Activation of
NF-κB was also impaired in DDX41-shRNA cells in response to either c-di-GMP or c-di-AMP
(<bold><xref ref-type="fig" rid="F5">Fig. 5c</xref></bold>). Together, our findings suggest that DDX41 is a
critical PRR for c-di-GMP and c-di-AMP mediated IFN induction, and that its absence generates a defect in
downstream STING-dependent signaling.
</p>
</sec>
...
</sec>
This paragraph, when parsed out using modified JATS XSLT transforms yields this text (with sentences of interest highlighted and tokenization from our text processing pipeline ):
DDX41 interacts and co-localizes with the STING adaptor ( Supplementary Fig. 4a,b ) to facilitate DNA ligand dependent signal transduction 17 . Introduction of either c-di-GMP or c-di-AMP into D2SC cells led to enhanced DDX41-STING complex formation ( Fig. 5a ). In DNA dependent signaling pathways, STING further binds to the downstream kinase TBK1 to activate the type I IFN response 18 , 19 . Both c-di-GMP and c-di-AMP when transfected into control D2SC cells activated the formation of a STING-TBK1 complex, however, c-di-GMP and c-di-AMP mediated activation of the STING-TBK1 complex was almost completely abrogated in DDX41-shRNA cells ( Fig. 5b ) . Consequently, c-di-GMP and c-di-AMP mediated activation of TBK1, IRF3 and the downstream type I IFN effector, STAT1 was impaired in DDX41-shRNA cells ( Fig. 5c and Supplementary Fig. 5 ). Activation of NF-κB was also impaired in DDX41-shRNA cells in response to either c-di-GMP or c-di-AMP ( Fig. 5c ). Together, our findings suggest that DDX41 is a critical PRR for c-di-GMP and c-di-AMP mediated IFN induction, and that its absence generates a defect in downstream STING-dependent signaling.
Note that the sentences to either side of the desired text either contain citations (as superscripted numbers in the text or <xref ref-type="bibr" rid="R17"> elements in the XML) or references to other figures in the document (also denoted by xref tags in the xml).
In this XML, the figure legends of interest occur in a completely separate part of the file (right at the end and separate from the other parts)
<floats-group>
...
<fig id="F5" orientation="portrait" position="float">
<label>Figure 5</label>
<caption>
<p>c-di-GMP and c-di-AMP require DDX41 for STING dependent signaling.
(<bold>a</bold>) Immunoprecipitation and immunoblot analysis of DDX41-
STING interactions is D2SC cells transfected with c-di-GMP or c-di-AMP
for 4 h. (<bold>b)</bold> Immunoprecipitation and immunoblot analysis of
STING-TBK1 interactions in control-shRNA or DDX41-shRNA D2SC cells
transfected as in <bold>a</bold>. (<bold>c</bold>) Immunoblot analysis of
TBK1, IRF3, p65 and STAT1 phosphorylations in control-shRNA, DDX41-
shRNA or STING-shRNA D2SC mDCs transfected with c-di-GMP, c-di-
AMP, Poly (I:C) or B-DNA for 4 h. Data are representative of at least two
independent experiments.
</p>
</caption>
<graphic xlink:href="nihms-410677-f0005"/>
</fig>
...
</floats-group>
This yields the following text:
Figure 5 c-di-GMP and c-di-AMP require DDX41 for STING dependent signaling. ( a ) Immunoprecipitation and immunoblot analysis of DDX41-STING interactions is D2SC cells transfected with c-di-GMP or c-di-AMP for 4 h. ( b ) Immunoprecipitation and immunoblot analysis of STING-TBK1 interactions in control-shRNA or DDX41-shRNA D2SC cells transfected as in a . ( c ) Immunoblot analysis of TBK1, IRF3, p65 and STAT1 phosphorylations in control-shRNA, DDX41-shRNA or STING-shRNA D2SC mDCs transfected with c-di-GMP, c-di-AMP, Poly (I:C) or B-DNA for 4 h. Data are representative of at least two independent experiments.
Ideally, what we would like would be a table of all subfigures (1a, 1b, 1c, … , 6f), with their location in the text (this could be described in any way that we could reconstruct easily) and the text of the sentences as shown.
Files for this example are listed here:
Note that the text file has a bunch of tokens that look like this: __s_A__ or __e_A__. These are ‘start’ and ‘end’ tags for formatting and links that we might find useful in the information extraction process. How we use these is open for discussion.
Results Statements and Figure Statements from example texts provided by Pathway Logic