amia2009
TRANSCRIPT
![Page 1: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/1.jpg)
04
/12
/20
23
1
UMLS-INTERFACE AND UMLS-SIMILARITY:
OPEN SOURCE SOFTWARE FOR MEASURING PATHS AND SEMANTIC SIMILARITY
Bridget McInnes
Ted Pedersen
Serguei Pakhomov
![Page 2: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/2.jpg)
2
OBJECTIVE
Develop tools to automatically compute the semantic similarity between two concepts in
the biomedical domain using measures originally developed for general English using the Unified Medical Language System (UMLS)
![Page 3: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/3.jpg)
3
MOTIVATION Clustering symptoms and disorders found in the text
of clinical reports for post marking medication safety and surveillance
Identification of patients for clinical studies
Improving the sensitivity of document retrieval of scientific journals and clinical reports
Development of terminologies and ontologies
Clustering of biomedical documents
Word sense disambiguation
![Page 4: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/4.jpg)
4
UNIFIED MEDICAL LANGUAGE SYSTEM
Knowledge representation framework Contains 3 Main components:
Metathesaurus Semantic Network SPECIALIST Lexicon
![Page 5: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/5.jpg)
5
METATHESAURUS
Semi-automatically integrates biomedical concepts from over a 100 controlled medical terminologies
Source vocabularies are organized based on their Atomic Unique Identifiers
Metathesaurus is organized based on their Concept Unique Identifier (CUI)
![Page 6: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/6.jpg)
6
CONCEPT UNIQUE IDENTIFIERS (CUIS)
AUI A15588749
Cold Temperature
MSH SNOMED-CT
AUI A3292554
Low Temperature
CUI C0009264
Cold Temperature
![Page 7: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/7.jpg)
7
CUI INFORMATION
The concepts (AUIs) from the source vocabularies may contain information about the concept such as its Definition Relation information between the concepts
The information from the AUIs can be obtained through their respective CUIs
![Page 8: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/8.jpg)
8
RELATIONS BETWEEN CUIS IN MSH
AUI A15588749
Cold Temperature
AUI A0123939
Temperature
MSH
is-a
CUI C0009264
Cold Temperature
CUI C0039476
Temperature
PAR/CHD (MSH)
![Page 9: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/9.jpg)
9
RELATIONS BETWEEN CUIS IN SNOMED-CT
AUI A2887140
Temperature
is-a
CUI C0009264
Cold Temperature
CUI C0039476
Temperature
PAR/CHD (SNOMED-CT)
SNOMED-CT
AUI A3292554
Low Temperature
![Page 10: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/10.jpg)
10
MULTIPLE RELATIONS
CUI C0009264
Cold Temperature
CUI C0039476
Temperature
PAR/CHD (SNOMED-CT)
CUI C0009264
Cold Temperature
CUI C0039476
Temperature
PAR/CHD (MSH)
![Page 11: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/11.jpg)
11
RELATION INFORMATION
AUI A15588749
Cold Temperature
AUI A0123939
Temperature
Relation
CUI C0009264
Cold Temperature
CUI C0039476
Temperature
Relation
MRHIER MRREL
![Page 12: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/12.jpg)
12
MRREL AND MRHIER MRHIER
Contains the full path to root relations between AUIs from each of the sources
is-a part-of
MRREL Contains the pairwise relations between CUIs Relations:
PAR/CHD RB/RN
It is possible to generate MRHEIR from MRREL except for the following sources: AIR MSH SNM2 USPMG OMS
![Page 13: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/13.jpg)
13
CUI VERSUS AUI HIERARCHY The benefit of using CUIs
Ability to obtain the relation information between concepts across sources
Ability to obtain the relation information between concepts using more than one type of relation:
PAR/CHD – parent/child (relation in MRHIER) RB/RN – narrower/broader SIB – sibling RL – concepts are similar or ‘alike’
The benefit of using AUIs Ability to obtain relation information (PAR/CHD) between
concepts in the same source very quickly incorporates tree positional information for sources such as
MSH
UMLS-Query by Shah and Musen, 2008
![Page 14: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/14.jpg)
14
UMLS-INTERFACE
Perl interface to the UMLS present locally in a MySQL database.
Its main purpose is to returns path information about CUIs using the relation information in MRREL All possible paths to the root Shortest path between two concepts
![Page 15: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/15.jpg)
15
UMLS-SIMILARITY
A suite of perl modules that implement a number of path-based semantic similarity measures to determine the similarity between two CUIs in the UMLS Measures are path-based because they rely on the
location of the concepts in a hierarchy The path information is obtained using UMLS-
Interface Semantic Similarity Measures:
Path measure Conceptual Distance (Rada, et. al, 1989) Leacock and Chodorow, 1998 Wu and Palmer, 1994 Nguyen and Al-Mubaid, 2006
![Page 16: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/16.jpg)
16
SEMANTIC SIMILARITY EXAMPLE
Path measure
1
where N = # links in the shortest path between the two concepts c1 and
c2
NSim(c1,c2) =
![Page 17: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/17.jpg)
17
SIMILARITY GIVEN SPECIFIED SOURCES
PAR (SNOMED-CT)
C0015385Limbs PAR
(SNOMED-CT)
C0229962Anatomic
Part
C0005898Body
Regions
PAR(MSH)
![Page 18: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/18.jpg)
18
SIMILARITY GIVEN SPECIFIED SOURCES
PAR (SNOMED-CT)
C0015385Limbs PAR
(SNOMED-CT)
C0229962Anatomic
Part
C0005898Body
Regions
PAR(MSH)
Similarity = 1/1
![Page 19: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/19.jpg)
19
SIMILARITY GIVEN SPECIFIED SOURCES
PAR (SNOMED-CT)
C0015385Limbs PAR
(SNOMED-CT)
C0229962Anatomic
Part
C0005898Body
Regions
PAR(MSH)
Similarity = 1/2
![Page 20: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/20.jpg)
20
SIMILARITY GIVEN SPECIFIED SOURCES
PAR (SNOMED-CT)
C0015385Limbs PAR
(SNOMED-CT)
C0229962Anatomic
Part
C0005898Body
Regions
PAR(MSH)
Similarity = 1/1
![Page 21: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/21.jpg)
21
SIMILARITY GIVEN SPECIFIED RELATIONS
RB(MSH)
C0015385Limbs
RB(MSH)
C0229962Anatomic
Part
C0005898Body
Regions
PAR(MSH)
![Page 22: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/22.jpg)
22
SIMILARITY GIVEN SPECIFIED RELATIONS
RB(MSH)
C0015385Limbs
RB(MSH)
C0229962Anatomic
Part
C0005898Body
Regions
PAR(MSH)
Similarity = 1/1
![Page 23: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/23.jpg)
23
SIMILARITY GIVEN SPECIFIED RELATIONS
RB(MSH)
C0015385Limbs
RB(MSH)
C0229962Anatomic
Part
C0005898Body
Regions
PAR(MSH)
Similarity = 1/2
![Page 24: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/24.jpg)
24
SIMILARITY GIVEN SPECIFIED RELATIONS
RB(MSH)
C0015385Limbs
RB(MSH)
C0229962Anatomic
Part
C0005898Body
Regions
PAR(MSH)
Similarity = 1/1
![Page 25: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/25.jpg)
25
FUNCTIONAL VALIDATION
Comparison with Previous Work: Pedersen, et al. 2007 Nguyen and Al-Mubaid, 2006 Caviedes and Cimino, 2004
![Page 26: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/26.jpg)
26
PEDERSEN, ET AL.
Semantic Similarity Measures Path Leacock and Chodorow, 1998
Source SNOMEDCT
Data 29 medical terms pairs Similarity determined by:
9 Medical Coders 3 Physicians
4 Point Scale 4 – practically synonymous 3 – related 2 – marginally related 1 - unrelated
Spearman’s Rank Correlation Coefficient
![Page 27: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/27.jpg)
27
COMPARISON WITH PEDERSEN, ET AL.
Semantic Similarity Measures Path Leacock and Chodorow, 1998
Source: SNOMED-CT from UMLS 2008AB
Relations: PAR/CHD
Comparison with human annotations Spearman Rank Correlation Coefficient
![Page 28: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/28.jpg)
28
COMPARISON WITH PEDERSEN, ET AL.
Measure Physician
Coder
path Pedersen, et. al.
0.36 0.51
UMLS-Similarity
0.35 0.50
Leacock and Chodorow
Pedersen, et. al. 0.35 0.50
UMLS-Similarity 0.35 0.50
![Page 29: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/29.jpg)
29
COMPARISON WITH PEDERSEN, ET AL.
Measure Physician
Coder
path Pedersen, et. al. 0.36 0.51
UMLS-Similarity 0.35 0.50
Leacock and Chodorow
Pedersen, et. al.
0.35 0.50
UMLS-Similarity
0.35 0.50
![Page 30: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/30.jpg)
30
NGUYEN AND AL-MUBAID
Semantic Similarity Measures Nguyen and Al-Mubaid, 2006 Leacock and Chodorow, 1998 Wu and Palmer, 1994 Path
Source: MSH Same Dataset created by Pedersen, et al.
Data 29 medical terms pairs Similarity determined by:
9 Medical Coders 3 Physicians
Spearman’s Rank Correlation Coefficient
![Page 31: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/31.jpg)
31
COMPARISON WITH NGUYEN AND AL-MUBAID
Semantic Similarity Measures Nguyen and Al-Mubaid, 2006 Leacock and Chodorow, 1998 Wu and Palmer, 1994 Path
Source: MSH from UMLS 2008AB
Relations: PAR/CHD
Comparison with human annotations Spearman Rank Correlation Coefficient
![Page 32: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/32.jpg)
32
COMPARISON WITH NGUYEN AND AL-MUBAID
Measure Physician
Coder
path Nguyen and Al-Mubaid
0.63 0.85
UMLS-Similarity 0.49 0.58
Leacock and Chodorow
Nguyen and Al-Mubaid
0.67 0.86
UMLS-Similarity 0.49 0.58
Wu and Palmer Nguyen and Al-Mubaid
0.65 0.79
UMLS-Similarity 0.45 0.54
Nguyen and Al-Mubaid
Nguyen and Al-Mubaid
0.67 0.86
UMLS-Similarity 0.45 0.55
![Page 33: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/33.jpg)
33
CAVIEDES AND CIMINO
Semantic Similarity Measure Conceptual Distance – Rada, et al.
Source: MSH Relations: PAR/CHD Data
10 medical terms pairs using following CUIs Digestive system disease: C0012242 Peptic esophagitis: C0014869 Psychotherapy: C0033968 Thirst: C0039971 Thoracic duct: C0039979
![Page 34: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/34.jpg)
34
COMPARISON WITH CAVIEDES AND CIMINO
Semantic Similarity Measures Conceptual Distance
Originally proposed by Rada, et. al., 1989
Source: MSH from UMLS 2008AB Relations: PAR/CHD
Comparison between the Conceptual Distance Scores
![Page 35: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/35.jpg)
35
COMPARISON WITH CAVIEDES AND CIMINO
CUI Pairs Caviedes and Cimino
UMLS-Similarity
C0012242-C0014869
3 3
C0012242-C0033968
5 5
C0033968-C0039971
6 6
C0012242-C0039971
7 7
C0012242-C0039979
7 6
C0033968-C0039979
8 9
C0014869-C0033968
8 8
C0014869-C0039971
10 10
C0014869-C0039979
10 11
C0039971-C0039979
10 11
![Page 36: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/36.jpg)
36
COMPARISON WITH CAVIEDES AND CIMINO
CUI Pairs Caviedes and Cimino
UMLS-Similarity
C0012242-C0014869
3 3
C0012242-C0033968
5 5
C0033968-C0039971
6 6
C0012242-C0039971
7 7
C0012242-C0039979
7 6
C0033968-C0039979
8 9
C0014869-C0033968
8 8
C0014869-C0039971
10 10
C0014869-C0039979
10 11
C0039971-C0039979
10 11
![Page 37: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/37.jpg)
37
RESULTS
The results show that UMLS-Similarity can be used to reproduce the results reported by:
Pedersen, et al.
Caviedes and Cimino
![Page 38: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/38.jpg)
38
RESULTS
The correlation results obtained by UMLS-Similarity and reported by Nguyen and Al-Mubaid vary Different versions of MSH were used to conduct
the experiment Possibly different mappings of the terms to CUIs
in MSH were used Information used by Nguyen and Al-Mubaid
comes directly from MSH which is located in MRHEIR and as PAR/CHD relations in MRREL It is not possible to generate MRHIER from MRREL
because the full path-to-root is a transitive closure of the pairwise PAR/CHD relations which does not hold true for MSH because a MSH concept may have different children depending on its tree position
![Page 39: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/39.jpg)
39
CONCLUSIONS
UMLS-Similarity Used to determine the similarity between two
concepts given a specified set of sources and relations
Contains the following similarity measures Path measure Conceptual Distance proposed Rada, et. al. 1989 Leacock and Chodorow, 1998 Wu and Palmer, 1994 Nguyen and Al-Mubaid, 2006
UMLS-Interface Used to obtain path information about a CUI
given a specified set of sources and relations
![Page 40: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/40.jpg)
40
FUTURE WORK
UMLS-Interface Improve the efficiency in which the path
information is stored
UMLS-Similarity Information Content Similarity Measures
Resnik, 1995 Jiang and Conrath, 1997 Lin, 1997
Relatedness Measures Patwardhan, 2003
![Page 41: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/41.jpg)
41
TAKE HOME MESSAGE #1
UMLS-Interface can used to extract path information about
a concept given a specified set of sources and relations.
![Page 42: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/42.jpg)
42
TAKE HOME MESSAGE #2
UMLS-Similarity can be used to compute the semantic similarity between two concepts given a
specified set of sources and relations.
![Page 43: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/43.jpg)
43
UMLS-Interface
http://search.cpan.org/dist/UMLS-Interface
UMLS-Similarity
http://search.cpan.org/dist/UMLS-Similarity
AVAILABILITY
![Page 44: Amia2009](https://reader035.vdocument.in/reader035/viewer/2022081602/55503f31b4c9058f768b48b0/html5/thumbnails/44.jpg)
44
THANK YOU
We would like to thank Kin Wah Fung, Olivier Bodenreider, Jan Willis Lan Aronson
The research was supported in parts by: Fellowships:
NLM Research Participation Program GAANN fellowship from US Dept. of Ed.
Grants IR01LM009623-01A2 from NIH, NLM