binding_17 extractions of "associated with"

[bgyori]

The binding_17 rule extracts complex formation events for A is associated with B sentences. However, is associated with seems to be usually used to draw an observational, statistical link between "distant" molecular players or higher level processes - things like Smoking is associated with lung cancer. The associated with pattern accounts for a lot of the extractions from binding_17 (~50%) and it turns out that these sentences overwhelmingly talk about associations that are not related to complex formation based on the samples we looked at.

Here are some examples of failure modes:

• Nogo-66 collapse inducing activity is associated with high-affinity binding to its receptors NgR1, which forms a complex with the transmembrane proteins LINGO1, and p75 or TROY.
• Oxidative damage to proteins by reactive oxygen species is associated with oxidative stress, aging, and age related diseases .
• Vasodilator stimulated phosphoprotein (VASP) is associated with FAs and is thought to have an important role in F-actin assembly.
• Increasing BMI and HOMA-IR values were significantly associated with lower eGFR.
• Our results showed that the polymorphisms rs12526453 and rs11066301 are significantly associated with the CAD risk in multiple populations.
• STAMP2 may be causally associated with IR, playing an important role in the regulation of glucolipid metabolism.
• Further immunohistochemical analysis suggested that TNBC in Malaysian women was strongly associated with EGFR, CK5/6 and c-KIT expression with high a Ki-67 proliferation index.

Finally, an example where associated with actually means complex formation (at least as far as I understand):

• Co-immunoprecipitation experiments clearly indicate that FLAG-T58A c-Myc was associated with HA-Rabring7 at the level similar to that of FLAG-wild-type c-Myc.

[MihaiSurdeanu]

@bgyori: we know "associated with" is ambiguous, but we're not sure how to disambiguate it. Any simple ideas?

[bgyori]

I have some ideas:

1. In a statistical sense simply removing this case of the rule would be a net improvement.
2. I would classify A associates with B as likely to refer to a physical interaction between two molecular entities and A is/was associated with B as unlikely to refer to a physical interaction.
3. Is named entity recognition and grounding able to influence what is ultimately extracted? If A and B are recognized as molecular entities (that is, proteins, protein families, complexes, chemicals) then extracting the complex formation event is warranted, otherwise the event should not be extracted.

[MihaiSurdeanu]

3 is a very good catch! We should do this! @myedibleenso @marcovzla: can you please make sure that Binding rules operate only proteins, families, or chemicals?

[MihaiSurdeanu]

@myedibleenso, @marcovzla: if the above idea doesn't work, let's go brute force and simply remove "associate" from the list of triggers. It seems too noisy.

[MihaiSurdeanu]

Another weird example for "associated with": Yields a binding event between TGF-beta and PAK2, even though the sentence doesn't support that::

[Evidence(reach, PMID22880011, {'cell_type': [u'cl:CL:0008019'], 'organ': None, 'tissue': None, 'location': None, 'cell_line': None, 'species': [u'taxonomy:4932']}, Pak2 is specifically activated by TGF-beta only in mesenchymal cells, as the result of phosphatidylinositol 3-kinase (PI3K) activation and may be associated with TGF-beta activation of Ras XREF_BIBR, XREF_BIBR, XREF_BIBR.)]

[MihaiSurdeanu]

@MihaiSurdeanu will try to clean binding_17 next.