Académique Documents
Professionnel Documents
Culture Documents
Motivation
The database of Genotypes and Phenotypes (dbGaP) is archiving the results of different Genome Wide Association Studies (GWAS). Phenotype variables are not harmonized across studies. Redundant phenotype identifiers for the same phenotype. dbGaP lacks semantic relations among its variables. Search on phenotypes is not accurate.
Goals
Standardize dbGaP information to allow accurate, reusable and Quick retrieval of information
id= phv00003636.v1 , Description= HEART: HYPERTENSIVE HEART DISEASE , name= FK414, version=1, Logical Max=--, Logical Minimum=--, unit=--, type=string
id= phv00008678.v3 , Description= CDI: HYPERTENSIVE HEART DISEASE , name= C334 , version=3, Logical Max=--, Logical Minimum=--, unit=--, type=text
N Alipanah,H Kim,L Ohno-Machado:Building an Ontology of Phenotypes for Existing GWAS Studies.Healthcare Informatics, Imaging and Systems Biology (HISB), 2012 IEEE Second International Conference on, vol., no., pp.111,27-28 Sept. 2012
id= phv00123020.v1, Description= CVD: self report of MD dx of cvd , name= cvd, version=1, value=Yes, No, Not assessed
Id= phv00123021.v1 , Description= CVD: self report of MD dx of cvd (missing recoded as no) , name=cvdx, version=1, value=Yes, No, Not assessed
Proposed Solution
Build
an information model
.
phv00123021.v1 phv00123020.v1
id= phv00008678.v3
Methods
I. String-based Variables Distance Calculation II. Semantic Hierarchy Extraction on Revised Clusters III. Classification and Ontology Creation IV. sdGaP Information Retrieval
1- Property Extraction Name, Description, Type, Unit, ,and (Max-Min) values 2- UMLS Expansion Expand Variable Description with MetaMap
4-
Build Distance Matrix Compute the Distance between every pair of Variables. Cluster based on Distance Matrix Variables with the same distance to other variables are clustered together.
5-
Build String-based distance matrix for variables in a single assigned cluster. Sub-cluster variables and calculate semantically relevant (similar) variables. Assign labels to sub-clusters based on the relevant UMLS Concept Unique Identifier. Perform re-clustering to find smaller group of relevant variables.
2.
3.
4.
2.
Extract corresponding sub class (PAR/CHD) hierarchy using the UMLS hierarchy table (MRREL table). Instantiate the phenotype variables to the UMLS CUIs. (Not for higher levels) Populate the related constraints in sdGaP
3.
4.
IV. Result
Dataset: Cleveland Family Study (CFS) with 5 data sets and 2,339 phenotype variables. (phs000284.v1.p1) Use Weka Tool for Xmean clustering. The X-mean clustering resulted in 35 clusters for relevant variables. Reorganized into 23 clusters by domain expert reviewers
Phv00122274.v1
Phv00122277.v1 Phv00122280.v1
Cardiovascular Disease
phv00123021.v1 phv00123020.v1
Phv00122281.v1
Conclusion
Extracting
Standard Reusable Information Model From UMLS Information Retrieval by Organizing Phenotype Variables and Instantiate them to Data Model
Improving
Limitation
Clustering
based on Distance Calculation is Semi-automated Computation. Instantiating variables to lower levels of hierarchy needs domain expert review. Only instances of lower level of hierarchies are considered in ontology building. For large data, distance calculation and clustering needs more advance algorithms.
Acknowledgement
Supported
by Grants