Support FamPlex

[jvwong]

Summary

Added support for FamPlex namespaced entities (family and complex) including 31 tests for FamPlex entities. I corrected some older test cases that, in fact, references families (e.g. Acc, p70S6K).

There are some additional changes and dependencies that should be addressed at some point (see Outstanding and Related issues)

Entity "type"?

The ENTRY_TYPE for famplex is currently set to 'entity'.

There are several alternatives to consider:

• Leave as entity
  • Most appropriate in factoid
  • Issues
    • Curation tool: entity doesn't appear to be an option; update required
    • Export to BioPAX and SBGN
      • SBGN: entity is labelled as SBGN node type 'unspecified entity' which is OK for now
      • BioPAX: entity is labelled as bp:PhysicalEntity but no external grounding is attached.
• Set as ggp
  • Easiest and doesn't require other changes to grounding UI, export formats
  • For family members, this may be reasonable but for complexes, it feels odd as they are not single genes or molecules.
• Set type as complex
  • factoid expects a complex to consist of > 1 components. However, this is complicated because (Famplex) complexes can consist of different combinations of components (e.g. proteins) some of which may be family members
  • factoid doesn't 'ground' parent complex
• Create new type complex/family
  • this starts to feel like entity anyways

Test results

Added famplex.json

Test cases added for each of the top 15 most frequently referenced entities:

"We found that the 15 most frequently-referenced FamPlex identifiers accounted for 50% of all FamPlex groundings (blue bars in Fig. 3d)"
Bachman, J. A., Gyori, B. M. & Sorger, P. K. FamPlex: a resource for entity recognition and relationship resolution of human protein families and complexes in biomedical text mining. BMC Bioinformatics 19, 248 (2018).

Results

I summarize the search accuracy, without and with FamPlex ( texts, data ):

FamPlex included	type	passes	failures	rate
No	search	649	105	13.9%
Yes	search	680	107	13.6%

Below, I report additional failing cases when Famplex data and tests are included:

text	expected.namespace	expected.id	actual.namespace	actual.id	organismOrdering	*rank
let-7	ncbi	266952	fplx	MIRLET7	[6239]	-1
UNC-40	ncbi	172233	ncbi	32757	[6239]	-1
lin-35	ncbi	172249	ncbi	55957	[6239]	-1

**rank indicates the rank order in search results; -1 means it was not returned at all

Refs #130

[jvwong]

Sample outputs:

biofactoid

app-ui (SBGN)

neo4j