Challenges of Term Recognition in Biology

1
Challenges of Term Recognition in Biology

• Sophia Ananiadou
• http//www.cs.salford.ac.uk/NLP.html
• University of Salford
• UK National Text Mining Centre

2
Outline

• Automatic term recognition in Biology
• ATR vs NER
• Term formation patterns
• Nested terms
• Term variation
• Acronyms
• Conclusion

3
Automatic Term Extraction

• One of the most crucial research topics in
  Biomedicine
• New terms (names of genes, proteins, drugs etc)
  are constantly created
• Existing resources are not sufficient
• over 280 databases in use (manually created and
  curated)
• controlled vocabularies are static repositories
• impossible to update manually terminologies due
  to dynamic nature of domain

4
Automatic Term Recognition

• ATR deals with spotting terms and their variants
  in texts and producing a list of candidate terms
• Recognition is the first step of term
  identification
• Classification (broad or fine grained) is the
  next step
• Last step is term integration with terminological
  resources, dictionaries, ontologies and databases
  

5
The whole picture
Term identification
Recognition
Classification
Integration
6
NER and ATR

• ATR recognition grouping of term variants
• NER recognition classification of occurrences
  of terms into specific classes e.g. gene,
  protein
• In NER the steps of recognition and
  classification are merged, a classified
  terminological instance is a named entity
• The tasks of ATR and NER share techniques but
  their ultimate goals are different
• ATR for resource building, lexica ontologies
• NER first step of IE, text mining

7
Naming in Biology

• Naming conventions in Biology
• Term formation guidelines from formal bodies e.g.
  Guidelines for Human Gene Nomenclature
• Are these sufficient? How often do we encounter
  ad-hoc names?
• Bride of sevenless (boss)
• Yotiao
• Term formation patterns
• Use of existing resources, narrow, widen or
  adjust the meaning of a wordform
• Use of general English words causes ambiguity
  was, not

8
Term formation patterns

• Modification of existing resources
• Affixation (transferase)
• Compounding (retinoic acid receptor)
• Acronyms (RAR retinoic acid receptor)
• Neologisms
• Foreign imports (cytomegalovirus)
• Numerals (9-cis retinoic acid)
• Symbols (Ca2 -calmodulin-dependent protein)
• Eponyms (Ewing sarcoma)

9
Inner structure of terms

• Majority of terms are multi-word units
• Restricted set of syntactic patterns
• Maximal vs nested term
• leukaemic T cell line Kit225
• Recognising the boundaries of multi-word terms
• In Genia, 1/3 of all nested terms appear more
  than once as nested, half of nested terms do not
  appear independently on the corpus
• Spotting nested terms on their own in corpus not
  sufficient
• Establishing semantic relationships among
  constituents (as a first step to building
  ontologies)

10
Term variation and usage of synonyms

• High correlation between degree of term variation
  and dynamic nature of domain
• Terminological variation is the ability to
  realise a concept in many ways
• A term variant is a synonym there is no change
  in meaning
• Variation occurs in controlled vocabularies and
  in texts
• Discrepancy of types of variation between the two

11
Types of variation

• Orthographic
• amino acid / amino-acid, oestrogen / estrogen
• Morphological
• cellular gene / cell gene / cell genes
• Lexical
• Carcinoma / cancer
• Structural
• Prepositional variants promoter of gene
• Coordinated variants adrenal glands and gonads
• Acronyms

12
Term variation part of ATR

• Enhancing performance through linguistic
  normalisation of variants
• Is it term coordination or conjunction?

13
Limitations (coordination)

• The recognition and extraction of coordinated
  terms is highly ambiguous
• Argument coordination (B and T cells) 90
• Head coordination (adrenal glands and gonads) 10
• Morphosyntactic clues are not sufficient as not
  systematic (argument coordination)
• Jun and Fos families
• mRNA and protein levels
• More background knowledge needed for the
  extraction and decoding of term coordinations

14
Coordination

• N1 and N2 PCP N3
• chicken and mouse stimulating factors
• chicken stimulating factor (N1 PCP N3)
• mouse stimulating factor (N2 PCP N3)
• dimerization and DNA binding domains
• dimerization domain (N1 N3)
• DNA binding domain (N2 PCP N3)

15
Acronyms

• Acronyms are a very productive type of term
  variation
• Acronym variation (synonymy)
• NF kappa B / NF kB / nuclear factor kappa B
• Acronym ambiguity (polysemy) even in controlled
  vocabularies
• GR for glucocorticoid receptor and glutathione
  reductase

16
Acronyms (2)

• Acronym patterns acronym formation structures
• Acronym definition patterns are syntactic
  patterns describing contexts where acronyms are
  introduced
• Linking acronyms with expanded forms (EFs)
• Acronym generation from EFs as part of term
  variation
• An issue is the selection of an optimal EF window
  for correct matching between acronym and EF

17
Acronyms (3)

• Most well known acronyms not defined in the
  corpus only 25 acronyms defined in corpus
• Need of acronym dictionaries to link acronyms
  with expanded forms
• Further improvement of acronym acquisition
  systems (90 precision, 70-80 recall) on
  syntactic / semantic information
• Acronym variation ambiguity still a challenge

18
Summary

• Handling of variations is an integral part of ATR
• Nested terms (recognition of boundaries
  semantic relationships among constituents)
• Unobserved variation generation (structural,
  spelling, acronym)
• Combining existing ontologies with ATR for finer
  and broader semantic classification and
  integration with databases

19
UK National Centre for Text Mining

• JISC award (2004-2007)
• Establish a high quality service provision for
  the UK bioscience community
• Identify the best of breed tools academia
  industry
• Types of services
• Facilitating access to tools, resources support
• Offering on-line use of resources and tools (also
  to guide and instruct)
• Offering a one-stop shop for complete, end-to-end
  processing
• We welcome collaboration. Contact
• S.Ananiadou_at_salford.ac.uk

20
References

• Frantzi, K., Ananiadou, S. Mima, H. (2000)
  Automatic recognition of multi-word terms.
  International Journal of Digital Libraries
  3115-130, Springer-Verlag
• Nenadic, G., Spasic, I., Ananiadou, S. (2002)
  Automatic acronym acquisition and management with
  domain specific texts, in Proc.LREC-3, Las
  Palmas, Spain, 2155-2162
• Nenadic, G., Spasic, I. Ananiadou, S. (2003)
  Terminology-Driven Mining of Biomedical
  Literature, in Journal of Bioinformatics, Vol.
  19(8), 938-943, Oxford University Press.
• Nenadic, G., Spasic, I. and Ananiadou, S. (2004)
  Mining the Biomedical Literature Whats in a
  Term? in Proceedings of International Joint
  Conference on Natural Language Processing
  (IJCNLP-04), Hainan Island, China, 247-254