TM-Based plugin - NER of gene names needs improvement

Rothamsted / knetbuilder

KnetBuilder data integration platform for building knowledge graphs. Previously known as ondex.

https://knetminer.com

MIT License

12 stars 11 forks source link

TM-Based plugin - NER of gene names needs improvement #10

Open KeywanHP opened 7 years ago

KeywanHP commented 7 years ago

Missing relation: TPP7 is a concept name of a gene but TM method doesn't map it to publication that contains OsTPP7 in its abstract.

https://www.ncbi.nlm.nih.gov/pubmed/16688177

KeywanHP commented 6 years ago

False positive relation between SSP gene and milling created. Ambiguous meaning of SSP:

Synthetic hexaploid wheat (T. turgidum ssp. dicoccoides x T. tauschii) as a source of favourable alleles for milling and baking quality traits.

KeywanHP commented 5 years ago

Three gene nodes in the network have preferred name "MFT" but only one was linked to trait "germination rate".

Evidence sentence: Recent studies in both Arabidopsis and wheat have uncovered a new role of MOTHER OF FT AND TFL1 (MFT) in seed germination. [PMID:24932489]

marco-brandizi commented 5 years ago

Does this depend on the TM plug-in only? Most importantly, does it make sense to consider one case at a time, or is it more sensible to assess overall scores like precision/recall?

KeywanHP commented 5 years ago

Precision/recall would be good if we had a gold standard. Until then I think we will need to go case by case and slowly build our own gold standard for more systematic evaluation in the future.

KeywanHP commented 3 years ago

A proper review of the text mining plugin is needed. I think it can be rewritten using cutting-edge NLP libraries.