Using gwas data for enhanced mendelian randomization studies
This presentation is the property of its rightful owner.
Sponsored Links
1 / 10

Using GWAS Data for Enhanced Mendelian Randomization Studies PowerPoint PPT Presentation


  • 228 Views
  • Uploaded on
  • Presentation posted in: General

Genetic Risk “G” (e.g., Risk Score, Candidate Genes). Phenotype/ Risk Factor “X” (e.g., Anxiety ). Outcome “Y” (e.g., CHD, Diabetes. Using GWAS Data for Enhanced Mendelian Randomization Studies. Stefan Walter ( [email protected] ) on behalf of:

Download Presentation

Using GWAS Data for Enhanced Mendelian Randomization Studies

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

Presentation Transcript


Using gwas data for enhanced mendelian randomization studies

Genetic Risk “G”

(e.g., Risk Score, Candidate Genes)

Phenotype/

Risk Factor “X”

(e.g., Anxiety)

Outcome “Y”

(e.g., CHD, Diabetes

Using GWAS Data for Enhanced Mendelian Randomization Studies

Stefan Walter

([email protected])

onbehalfof:

Laura Kubzansky, Maria Glymour, Eric J TchetgenTchetgen, Liming Liang, Shun-Chiao Chang, Eric Rimm, Marilyn Cornelis, KarestanKoenen, and Ichiro Kawachi


Outline

Outline

  • Setting

  • Projects:

    • Anxiety, Depression, SocialTies

    • Results

  • Methods and Instruments:

    • Using the Genetic Risk Score as an Instrument

    • Leave One Out Approach

    • Instrumental Inequality Tests

  • Future:

    • including interactions in score (GxG, GxE)


Setting

Setting

  • NHS I

    • n = 7000 (NHST2D, NHSCGEMS, NHSCHD, NHSKS)

    • age at blooddraw = 59.5 (SD 8)

  • HPFS

    • n = 4125 (HPFST2D, HPFSCHD, HPFSKS)

    • age at blooddraw = 62.2 (SD 9)

      Genotypedondifferent chips (Illumina, Affy), HapMap and 1000g imputationavailable.


Projects

Projects

  • Anxiety:

    • Crown Crisp Experimental Index (phobic anxiety)

  • Depression

    • long-term composite depressive symptom score (1992-2006).

  • Social Ties

    • social isolation and ability to connect with others (continuously married versus otherwise).


Results

Results

  • Depression and Anxiety:

    • No evidence of a geneticinstrumentfromadditiveinternal GWA risk score, candidate genes, orexternal GWA risk score (R2 < 0.1%, mostlynotsignificant)

      Butinconsistentresultsfrom chip heritability (GCTA):

      NHS_T2D, Affy 6.0, 0.214349

      NHS_CGE, Illumina 550k, 0.062074

      NHS_CHD, Affy 6.0, 0.305586

      NHS_KS, Illumina 610Q, 0.000001

      HPFS_T2D, Affy 6.0, 0.214349

      HPFS_CHD, Affy 6.0,0.200122

      HPFS_KS, Illumina 610k, 0.046558


Using gwas data for enhanced mendelian randomization studies

Genetic Risk “G”

(e.g., Risk Score, Candidate Genes)

Phenotype/

Risk Factor “X”

(e.g., BMI)

Outcome “Y”

(e.g., Anxiety, Depression, Social Ties)

  • Advantage:

    • Knowngeneticrelationshipfor BMI (Speliotes et al. Association analyses of 249,796 individuals reveal 18 new loci associated withbodymassindex).

    • Replicates in NHS/HPFS explaining 2% of variance.

  • Challenge:

    • Analysiswithinnested case-control (NCC) samples (allbut KS) requires IP weightingtoallowunbiasedinferencebasedonthe original samplingpopulation.

    • Currently, we are recreatingtherisk sets basedonthepublishedmatchingcriteriausedto derive the NCC.


Methods and instruments 1

Methods and Instruments (1):

  • Genetic Risk Score applying allele scoring (Purcell et al., 2009):

    • Derived from internal GWAS

      • Approach: running GWAS in 7 samples, meta-analyzing 6 and scoring in set number 7. Iterative leave one out procedure and subsequent meta-analysis of results.

    • Derived from published GWAS (Speliotes et al, 2010; Demirkan et al., 2010)

    • Derived from Candidate Genes


Methods and instruments 2

Methods and Instruments (2):

Glymour, TchetgenTchetgen, Robins., CredibleMendelianRandomizationStudies: Approaches for Evaluating the Instrumental Variable Assumptions, Am J Epi2012

Genetic IVs cannot be proven to be valid. They can sometimes be shown to be invalid, although these tests generally rely on additional assumptions.

  • Four empirical approaches to (in)validation:

    • Leverage prior causal assumptions regarding the confounding of the phenotype-outcome association: 4 equivalent versions of this test.

    • Identify factors that modify the genotype-phenotype association and compare the IV effect estimate across values of the modifier.

    • Instrumental inequality tests: applicable only when the causal phenotype is known to be categorical.

    • Over-identification tests with multiple IVs. Other genes or even polymorphisms of the same gene might provide additional IVs.

      *Instrumental Inequality Test

    • Macros available (R, SAS) fordichtomousoutcome, dichotomousphenotype, and ploytomousinstruments


Future

Future

Testing and includingGxG, GxE

  • KnownProtein x Proteininteractionstoinformunderlying SNP x SNP interactions. (anxiety)

  • Investigategenderdifferences. (anxiety)

  • QuantileRegressiontoidentify and quantifyinteractionswith (unknown) environmentalfactors.

  • Construct separate scores based on presumed mechanism (e.g., appetite, adipogenesis, cardio-pulmonary fitness) and apply over-identification tests.

  • GetaccesstotheHealthRetirementStudygenetic data fromdbGaPtoincreasesamplesizeforpsycho-social phenotypes.


Acknowledgements contact

Acknowledgements & Contact:

NIH/NIHM 1RC4MH092707 (L Kubzansky)

And the entire team:

Laura Kubzansky, Maria Glymour, Eric J TchetgenTchetgen, Liming Liang, Shun-Chiao Chang, Eric Rimm, Marilyn Cornelis, KarestanKoenen, and Ichiro Kawachi

Please feel free to contact us, we are happy to share thoughts, code, macros, etc.

Stefan: [email protected]

Laura: [email protected]

Maria: [email protected]


  • Login