may 30, 2005bionlp lexical issues in anatomical ontologies sarah luger based on work with stuart...

19
May 30, 2005 BIONLP Lexical Issues in Anatomical Ontologies Sarah Luger www.xspan.org based on work with Stuart Aitken and Bonnie Webber

Upload: elfreda-nichols

Post on 23-Dec-2015

214 views

Category:

Documents


0 download

TRANSCRIPT

May 30, 2005 BIONLP

Lexical Issues in Anatomical Ontologies

Sarah Lugerwww.xspan.org

based on work with Stuart Aitken and Bonnie Webber

May 30, 2005 BIONLP

XSPAN Objectives• To support access to

tissue-based genetic information that is indexed by anatomy across species, via anatomy ontology integration.

– homology relationships– function similarities– lineage relationships– cell types

May 30, 2005 BIONLP

A Mapping Problem

• Mouse Tail

May 30, 2005 BIONLP

A Mapping Problem

• C. Elegans Tail

• Mouse tail to C. elegans tail Same name, different function

May 30, 2005 BIONLP

A Mapping Problem• In mouse:embryo/ organ system/ sensory organ/ eye/ optic

stalk/ optic nerve

• In drosophila:larva/ larval organ system/ larval nervous system/

larval central nervous system/ larval brain/ medulla anlage/ optic nerve

• Mouse optic nerve to drosophila optic nerve– Same name, same function– The ontologies show different paths.

May 30, 2005 BIONLP

A Mapping Problem

• Identify and exploit relationships between tissues across the key model organisms given:

– Mouse: 3559 anatomical parts– Drosophila: 506 anatomical parts– C. Elegans: 242 anatomical parts– Zebrafish: ~1200 anatomical parts– Human: 2328 anatomical parts

• Can their terminologies and anatomical ontogolies suggest what parts may be similar (homologous)?

May 30, 2005 BIONLP

Lexical Analysis

• Language processing:normalize, remove stop words, stemmed and lemmatized and then

treated as unordered set

• Use more than the leaf label– Context, terms are not unique– Important terms spread across path

• The last term is weighted more than all prior terms combined. This filters out some “garbage” when comparing similar root-to-leaf paths with vastly different levels of specificity.

May 30, 2005 BIONLP

Weighting Example• EMAPA.18237:

Mouse/ embryo/ organ system/ sensory organ/ eye/ lens/ capsule

0 1 2 4 8 16 32

• EHDAA.9045: 100% match

Human/ embryo/ organ system/ sensory organ/ eye/ lens/ capsule

0 1 2 4 8 16 32

• EHDAA.4727: 87% similar or 55/63

Human/ embryo/ organ system/ sensory organ/ eye/ lens vesicle/capsule

0 1 2 4 8 8 32

May 30, 2005 BIONLP

Weighting Example

1. EHDAA.9045: Human/ embryo/ organ system/ sensory organ/ eye/ lens/ capsule

2. EMAPA.18979: Mouse/ embryo/ organ system/ visceral organ/ reproductive system/ female/ ovary/ capsule

3. EMAPA.18679:

Mouse/ embryo/ organ system/ visceral organ/ renal urinary system/ metanepheros/ excretory component/ cortex/ capsule

May 30, 2005 BIONLP

Methodology• Use a similarity threshold to limit the number of results.

• Resultant pairs have one to many mappings:EMAPA: 18237 EHDAA.9045EMAPA: 18237 EHDAA.4727

• Tissue pairs then structurally assessed by taking the ontologies as graphs with directed but unlabeled edges.– Examine intra-species relationships– Check relative positions of the terms between species– Judge comparisons as contradictory, supported, or no evidence.

May 30, 2005 BIONLP

Mouse-Human Results

Threshold Matched Pairs Support

100% 1223 90.4%

96% 1763 84.7%

94% 3037 92.8%

This improvement follows from the larger set of lexical matches having more common terms near the top of the ontology, which permits paths that were previously being rejected.

May 30, 2005 BIONLP

Mouse-Human Results

Mappings between human and mouse terms, where the x-axis represents thresholds of lexical similarity.

Mapping Compatibility and Mapping Scores

0

10

20

30

40

50

60

70

80

90

100

100% 98% 96% 94% 92%

Human-Mouse (%)

Human-Mouse (mapping score)

MappingScore

7.0

MappingCompatibility (%)

0

May 30, 2005 BIONLP

Human-Drosophila Results

• Fewer matches so we chose lexical similarity levels of 75%, 66% and 50%.

• Some contradictory mappings appear at 50%.

• The proportion of compatible mappings drops as the comparison threshold is reduced.

• An increase in mapping score is again observed.

Mapping Compatibility and Mapping Scores

0

10

20

30

40

50

60

70

80

90

100

75% 66% 50%

Human-Drosophila (%)

Human-Drosophila(mapping score)

MappingCompatibility (%)

7.0

0

Mapping Score

May 30, 2005 BIONLP

Future Work

• Different name, same function– Introduces the need for synonym tables.

• Working with an anatomist to include synonyms in the mouse anatomy.– Zebrafish is also working on this.

• Discover more differences in ontology terms and free-text terms.

May 30, 2005 BIONLP

Ontologies, Mapping and Synonyms

drosophilaisa whole organism part of organ system isa reproductive system isa male reproductive

system part of testis

C. elegans part of organ system isa sex-associated system isa male-associated system isa male reproductive system isa male gonad

gonad

testis

May 30, 2005 BIONLP

COBrA: Ontology Mapping

May 30, 2005 BIONLP

COBrA: A GO editor

hyperlink

GO termdata

search

May 30, 2005 BIONLP

COBrA: A GO editor

May 30, 2005 BIONLP

Structural Analysis