Lecture Integrated Logic Systems

1
Lecture Integrated Logic Systems

• The Semantic Web and Ontology term extraction
  from research articles
• Andreas Doms
• adoms_at_biotec.tu-dresden.de

2
Bioinformatics and the Semantic Web

• NSF and EUs strategic research workshop found
  that bioinformatics could play the role for the
  semantic web, which physics played for the web.
• Why?
• Masses of information
• Data is public
• Data is online
• Data (more and more often) published in XML
• Data standards are accepted and actively
  developed by bioinformatics community
• Much valuable information scattered (as
  production cheap and hence often not centralised)
• Systems integration and interoperation prime
  concern
• Prediction In the not too distant future many
  tools and databases will be accessible as web
  services

3
The Web

• A great success story, but
• its the web for humans, not machines

4
Example Pubmed

• gt12.000.000 literature abstracts
• Great resource if one knows what one is looking
  for
• Kox1 has 17 hits
• But diabetes will produce gt200.000
• Often need to automatically process abstracts

5
Results of PubMed

• Lorenz P, Transcriptional repression mediated by
  the KRAB domain of the human C2H2 zinc finger
  protein Kox1/ZNF10 does not require histone
  deacetylation.Biol Chem. 2001 Apr382(4)637-44.
• Fredericks WJ. An engineered PAX3-KRAB
  transcriptional repressor inhibits the malignant
  phenotype of alveolar rhabdomyosarcoma cells
  harboring the endogenous PAX3-FKHR oncogene.Mol
  Cell Biol. 2000 Jul20(14)5019-31....

6
Results of PubMed

• ??????????????????????????????????????????????????
  ??????????????????????????????????????????????????
  ??????????????????????????????????????????????????
  ?????????????????????????????????????
• ??????????????????????????????????????????????????
  ??????????????????????????????????????????????????
  ??????????????????????????????????????????????????
  ?????????????????????????????????????????????????
  ????????????????....

7
Results of PubMed

• ??????????????????????????????????????????????????
  ??????????????????????????????????????????????????
  ??????????????????????????????????????????????????
  ?????????????????????????????????????????????????
  ??????????????????????????????
• ??????????????????????????????????????????????????
  ??????????????????????????????????????????????????
  ??????????????????????????????????????????????????
  ?????????????????????????????????????????????????
  ??????????????????????????????
• ...

8
GeneOntology

• Biologists have recognised the problem of
  semantic inter-operability between disparate
  information sources
• GeneOntology (GO) is effort to provide common
  vocabulary for molecular biology
• GO has gt19.000 terms in three branches
  function, process, localisation

9
Ontology Merging

• Big research efforts are being made in the field
  of the Semantic Web
• One problem here is the convergence of ontology
  from different sources
• Research group A develops an ontology for
  biological processes in mice
• Research group B develops an ontology for
  biological processes in human
• Many of the concept will describe the same
  process only in two different organisms

10
Ontology Merging

• How can we merge those two ontologies?
• How to automatically align concepts or subtrees?
• Matching of the concepts names?
• Graph matching of the descendants of the concept?
• Domain specific ontologies vs. one global
  ontology?

11
Browser

• Difference between WWW and the Semantic Web?
• How to browse the Semantic Web?
• Why does Google work so fine?
• Why is searching the in the Semantic Web
  different?

12
Browser

• Difference between WWW and the Semantic Web?
• Plain text or HTML with plain text vs. Semantic
  Facts and Web Services
• How to browse the Semantic Web?
• You want direct answers to your concrete
  questions
• Why does Google work so fine?
• Answers to be found in one single document,
  citations to good resources give heuristic
• Why is searching the in the Semantic Web
  different?
• Reasoning over facts, new information spread over
  several resources in the web

13
From HTML to XML

• The Semantic Web needs fact. (Semantically rich
  information)
• A resource A in the web can have an ltAuthorgt
• The ltAuthorgt is a resource B itself (personal
  webpage)
• Some fact base says that B ltcreatedgt A
• Most knowledge in the web is in plain text
• Only very slowly information will be created as
  semantic fact by web authors
• What about the currently available millions of
  sites?

14
Semantic Annotation

• Natural language is very ambiguous
• Try to extract fact from natural language texts
• First problem find and disambiguousize entities
• Use domain specific ontologies on domain specific
  texts

15
Example

• PubMed Artikel Nummer 7638186In this report we
  demonstrate that IL-12 induces tyrosine
  phosphorylation of a recently identified STAT
  family member, STAT4, and show that STAT4
  expression is regulated by T-cell activation.

tyrosine phosphorylation of STAT protein(is a
GOTerm!!!)
16
Term Extraction

• Sequence alignment algorithms very successful
  used in bioinformatics
• Can it be used for term extraction?

17
Term Extraction
18
Term Extraction

• No uniform vocabulary is used
• GeneOntology is an attempt to unify language
  molecular biology GOTerms
• isomerase activity
• cis-trans isomerase activity
• peptidyl-prolyl cis-trans isomerase activity
• cyclophilin-type peptidyl-prolyl cis-trans
  isomerase activity

19
The GeneOntology
Gene Ontology
Molecular function
Biological process
Cellular Component















20
Example

• Query for levamisole inhibitor retrieves more
  than 100 hits in PubMed.
• Which enzymatic functions does levamisole
  inhibitor have?
• Query for levamisole inhibitor enzymatic activity
  retrieves only 5 hits.
• Many relevant articles missing.

21
(No Transcript)
22
Transferase 8 Kinase 6 Hydrolase
58 Oxidoreductase 2 Lyase 1
23
Phosphofructokinase
24
In PubMed this articel is at position 84 listed
25
Levamisole direclty inhibits tumor
phosphofructokinase
26
Term Extraction

• Cant we use Sequence alignment for the term
  extraction?
• ATAGCTGCTAGCTAGATGTACTAGCATCGT
• GCATGTAGGCATC

27
Term Extraction

• What is similar to natural language
  texts?ATAGCTGCTAGCTAGATGTACTAGCATCGT
  GC---ATGTAG--GCATC

28
The idea

• PubMed Articel 7638186 Interleukin 12 induces
  tyrosine phosphorylation and activation of STAT4
  in human lymphocytes.Interleukin 12 (IL-12) is
  an important immunoregulatory cytokine whose
  receptor is a member of the hematopoietin
  receptor superfamily. We have recently
  demonstrated that stimulation of human T and
  natural killer cells with IL-12 induces tyrosine
  phosphorylation of the Janus family tyrosine
  kinase JAK2 and Tyk2, implicating these kinases
  in the immediate biochemical response to IL-12.
  Recently, transcription factors known as STATs
  (signal transducers and activators of
  transcription) have been shown to be tyrosine
  phosphorylated and activated in response to a
  number of cytokines that bind hematopoietin
  receptors and activate JAK kinases. In this
  report we demonstrate that IL-12 induces tyrosine
  phosphorylation of a recently identified STAT
  family member, STAT4, and show that STAT4
  expression is regulated by T-cell activation.
  Furthermore, we show that IL-12 stimulates
  formation of a DNA-binding complex that
  recognizes a DNA sequence previously shown to
  bind STAT proteins and that this complex contains
  STAT4. These data, and the recent demonstration
  of JAK phosphorylation by IL-12, identify a rapid
  signal-transduction pathway likely to mediate
  IL-12-induced gene expression.

29
The idea

• PubMed Articel 7638186 Interleukin 12 induces
  tyrosine phosphorylation and activation of STAT4
  in human lymphocytes.Interleukin 12 (IL-12) is
  an important immunoregulatory cytokine whose
  receptor is a member of the hematopoietin
  receptor superfamily. We have recently
  demonstrated that stimulation of human T and
  natural killer cells with IL-12 induces tyrosine
  phosphorylation of the Janus family tyrosine
  kinase JAK2 and Tyk2, implicating these kinases
  in the immediate biochemical response to IL-12.
  Recently, transcription factors known as STATs
  (signal transducers and activators of
  transcription) have been shown to be tyrosine
  phosphorylated and activated in response to a
  number of cytokines that bind hematopoietin
  receptors and activate JAK kinases. In this
  report we demonstrate that IL-12 induces tyrosine
  phosphorylation of a recently identified STAT
  family member, STAT4, and show that STAT4
  expression is regulated by T-cell activation.
  Furthermore, we show that IL-12 stimulates
  formation of a DNA-binding complex that
  recognizes a DNA sequence previously shown to
  bind STAT proteins and that this complex contains
  STAT4. These data, and the recent demonstration
  of JAK phosphorylation by IL-12, identify a rapid
  signal-transduction pathway likely to mediate
  IL-12-induced gene expression.

30
The idea

• Lets find tyrosine phosphorylation of STAT
  protein
• In this report we demonstrate that IL-12 induces
  tyrosine phosphorylation of a recently identified
  STAT family member, STAT4, and show that STAT4
  expression is regulated by T-cell activation.
  Gaps and one missing words is the word
  protein really needed for the correct
  extraction?

31
Local alignment for term extraction

• Like in sequence alignment for protein sequence
  the algorithm aligns words from the GOTerm
  against words from free text.
• Gaps in the GOTerm are allowed
• Deletion of words from the term are allowed
• Mutations are allowed in the form of stemmed
  word forms
• Activity,active ?activ (word stem)

32
Local alignment for term extraction

• What to do if a word of a GOTerm is missing in
  the text but the concept is clearly mentioned in
  the sentence.
• Do we align against the full text or against
  paragraphs or sentences?
• Currently the best results are being made with
  per sentence alignment
• How to penalize the gaps?

33
Local alignment for term extraction

• The current implementation
• Takes one articles abstract
• Tokenizes it into sentences
• For each potential term to be found in the
  sentence it aligns it with the sequence of word
  from the text (sentence and term are tokenized
  before using the same tokenizer)
• The scoring function (to compare two words) uses
  the concept of information content to calculate
  the gain or the penalty (for mismatches)
• Each term has an information content which is the
  sum of the information contents of ist word
• If all words of a term could be found (in the
  correct order, and there are no gaps) the term is
  matched 100

34
Extraction Examples

• Grammatical roots. (Terms are used in a modified
  version because of English grammar rules.)
• PMID 7744799 The protein products of this gene
  contain the basic-helix-loop-helix motif
  characteristic of a large family of transcription
  factors that bind to the canon ical DNA sequence
  CANNTG as protein heterodimers.
• is mentioning the GO term transcription factor
  binding (GO0008134), but stemming is needed here
  to identify bind as binding.

35
Extraction Examples

• Insertions. (Examples of insertion into the
  original term without changing the meaning are
  possible)
• PMID 7578980 Primed monocytes transcribed TNF
  mRNA at a higher rate than freshly isolated
  monocytes upon activation with LPS. (monocyte
  activation (GO0042117)) and, with even longer
  insertion
• PMID 7612661 Although all nm23 proteins contain
  nucleoside diphosphate (NDP) kinase activity, it
  has not been established that the enzyme activity
  mediated the various functions of nm23 proteins.
  (protein kinase activity (GO0004672)).
• These examples suggest to allow insertions into
  the terms. The insertions are words which do not
  invalidate the meaning of the original term.

36
Extraction Examples

• Hyphenated compounds vs. spaced words
• The use of hyphenated compounds and spaced words
  is not allways consistent Terms like
  thioredoxin-disulfide reductase activity
  (GO0004791) occur with and without the hyphen
  between the first two words.

37
Extraction Examples

• Alternatives. The term small-molecule carrier or
  transporter (GO0005468) can be mentioned in the
  form small-molecule carrier or small-molecule
  transporter.
• Endonuclease activity, active with either ribo-
  or deoxyribonucleic acids and producing
  5-phosphomonoesters (GO0016893) most likely
  will be used without the complementary subclause
  after the comma, although omitting it without
  reference to the context can make it ambiguous.
  1,239 terms (7.3) contain one or more commas.

38
Extraction Examples

• Square brackets.
• Terms have brackets in their names like
  methionine synthase reductase activity
  (GO0030586).

39
Extraction Examples

• Sensu extension.
• For terms like structural constituent of chorion
  (sensu Insecta) (GO0005213), it is unlikely that
  the author mentions them with the suffix (sensu
  Insecta), as it will be clear from the abstracts
  context that the topic is about insects.
• In GO, 438 terms have such a suffix.

40

• Comma-separated descriptive subclauses.
• In the GO version of March 2004, 799 terms have
  commas followed by a space, e.g. hydrolase
  activity, acting on acid anhydrides, acting on
  GTP, involved in cellular and subcellular
  movement (GO0050800). In the current GO version,
  this syntactical concept is used to encode
  relations other than is a or part of.

41
Frequent words in the Gene Ontology
42
Dynamic programming
43
Idea of dynamic programming
44
Backtracking
45
(No Transcript)
46
(No Transcript)
47
Other approaches

• Regular Expression
• Edit distances (?)
• Machine learning
• Linguistic analysis

48
Tutorial today

• Prepare your presentation at a machine in the
  lab.
• Each student must comment on the project