R Packages for Genome-Wide Association Studies

1
R Packages for Genome-Wide Association Studies

• Qunyuan Zhang
• Division of Statistical Genomics
• Statistical Genetics Forum
• March 10,2008

2
What is R ?

• R is a free software environment for statistical
  computing and graphics.
• Run s on a wide variety of UNIX platforms,
  Windows and MacOS (interactive or batch mode)
• Free and open source, can be downloaded from
  cran.r-project.org
• Wide range of packages (base contributed),
  novel methods available
• Concise grammar good structure (function, data
  object, methods and class)
• Help from manuals and email group
• Slow, time and memory consuming (can be overcome
  by parallel computation, and/or integration with
  C)
• Popular, used by 7080 statisticians

3
R Task Viewshttp//cran.r-project.org/web/views/
4
Statistical Genetics Packages in
Rhttp//cran.r-project.org/web/views/Genetics.htm
l

• Population Genetics genetics (basic), Geneland
  (spatial structures of genetic data), rmetasim
  (population genetics simulations), hapsim
  (simulation), popgen (clustering SNP genotype
  data and SNP simulation), hierfstat (hierarchical
  F-statistics of genetic data), hwde (modeling
  genotypic disequilibria), Biodem
  (biodemographical analysis), kinship (pedigree
  analysis), adegenet (population structure), ape
  apTreeshape (Phylogenetic and evolution
  analyses), ouch (Ornstein-Uhlenbeck models),
  PHYLOGR (simulation and GLS model), stepwise
  (recombination breakpoints)
• Linkage and Association gap (both population
  and family data, sample size calculations,
  probability of familial disease aggregation,
  kinship calculation, linkage and association
  analyses, haplotype frequencies) tdthap (TDT for
  haplotypes, powerpkg (power analyses for the
  affected sib pair and the TDT design),hapassoc
  (likelihood inference of trait associations with
  haplotypes in GLMs), haplo.ccs (haplotype and
  covariate relative risks in case-control data by
  weighted logistic regression), haplo.stats
  (haplotype analysis for unrelated subjects),
  tdthap (haplotype transmission/disequilibrium
  tests), ldDesign (experiment design for
  association and LD studies), LDheatmap (heatmap
  of pairwise LD),. mapLD (LD and haplotype
  blocks), pbatR (R version of PBAT), GenABEL
  SNPassoc for GWAS
• QTL mapping for the data from experimental
  crosses bqtl (inbred crosses and recombinant
  inbred lines), qtl (genome-wide scans),
  qtlDesign (designing QTL experiments power
  computations), qtlbim (Bayesian Interval QTL
  Mapping)
• Sequence Array Data Processing seqinr,
  BioConductor packages

5
GenABELAulchenko Y.S., Ripke S., Isaacs A., van
Duijn C.M. GenABEL an R package for genome-wide
association analysis. Bioinformatics. 2007,
23(10)1294-6.

• GenABEL genome-wide SNP association analysis
• a package for genome-wide association analysis
  between quantitative or binary traits and
  single-nucleotides polymorphisms (SNPs).
• Version 1.3-5
• Depends R ( 2.4.0), methods, genetics,
  haplo.stats, qvalue, MASS
• Date 2008-02-17
• Author Yurii Aulchenko, with contributions from
  Maksim Struchalin, Stephan Ripke and Toby Johnson
  
• Maintainer Yurii Aulchenko lti.aoultchenko at
  erasmusmc.nlgt
• License GPL ( 2)
• In views Genetics
• CRAN checks GenABEL results

6
GenABEL Data Objects
phdata phenotypic data (data frame)
gtdata genotypic data (snp.data-class)

• gwaa.data-class

• 2-bit storage
• 0 00
• 1 01
• 2 10
• 11
• Save 75

load.gwaa.data(phenofile "pheno.dat", genofile
"geno.raw)
convert.snp.text() from text file (GenABEL
default format) convert.snp.ped() from Linkage,
Merlin, Mach, and similar files convert.snp.mach()
from Mach format convert.snp.tped() from PLINK
TPED format convert.snp.illumina() from
Illumina/Affymetrix-like format
7
GenABEL Data Manipulation

• snp.subset() subset data by snp names or by QC
  criteria
• add.phdata() merge extra phenotypic data to the
  gwaa.data-class.
• ztransform() standard normalization of
  phenotypes
• rntransform() rank-normalization of phenotypes
• npsubtreated() non-parametric adjustment of
  phenotypes for medicated subjects

8
GenABEL QC Summarization

• summary.snp.data() summary of snp data (Number
  of observed genotypes, call rate, allelic
  frequency, genotypic distribution, P-value of HWE
  test
• check.trait() summary of phenotypic data and
  outlier check based on a specified p/FDR cut-off
• check.marker() SNP selection based on call rate,
  allele frequency and deviation from HWE
• HWE.show() showing HWE tables, Chi2 and exact
  HWE P-values
• perid.summary() call rate and heterozygosity per
  person
• ibs() matrix of average IBS for a group of
  people a given set of SNPs
• hom() average homozygosity (inbreeding) for a
  set of people, across multiple markers

9
GenABEL SNP Association Scans

• scan.glm() snp association test using GLM in R
  library
• scan.glm((yx1x2CRSNP", family
  gaussian(), data, snpsubset, idsubset)
• scan.glm((yx1x2CRSNP", family binomial
  (), data, snpsubset, idsubset)
• scan.glm.2D() 2-snp interaction scan
• Fast Scan (call C language)
• ccfast() case-control association analysis by
  computing chi-square test from 2x2 (allelic) or
  2x3 (genotypic) tables
• emp.ccfast() Genome-wide significance
  (permutation) for ccfast() scan
• qtscore() association test (GLM) for a trait
  (quantitative or categorical)
• emp.qtscore() Genome-wide significance
  (permutation) for qscaore() scan
• mmscore() score test for association between a
  trait and genetic polymorphism, in samples of
  related individuals (needs stratification
  variable, scores are computed within strata and
  then added up)
• egscore() association test, adjusted for
  possible stratification by principal components
  of genomic kinship matrix(snp correlation matrix)

10
GenABEL Haplotype Association Scans

• scan.haplo() haplotype association test using
  GLM in R library
• scan.haplo.2D() 2-haplotype interaction scan
• (haplo.stats package required)
• Sliding window strategy
• Posterior prob. of Haplotypes via EM algorithm
• GLM-based score test for haplotype-trait
  association (Schaid DJ, Rowland CM, Tines DE,
  Jacobson RM, Poland GA. 2002. Score tests for
  association of traits with haplotypes when
  linkage phase is ambiguous Am J Hum Genet 70
  425-434. )

11
GenABEL GWAS results from scan.glm,
scan.haplo, ccfast, qtscore, emp.ccfast,emp.qtscor
e

• scan.gwaa-class
• Names snpnames list of names of SNPs tested
• P1df p-values of 1-d.f. (additive or allelic)
  test for association
• P2df p-values of 2-d.f. (genotypic) test for
  association
• Pc1df p-values from the 1-d.f. test for
  association between SNP and trait the statistics
  is corrected for possible inflation
• effB effect of the B allele in allelic test
• effAB effect of the AB genotype in genotypic
  test
• effBB effect of the BB genotype in genotypic
  test
• Map list of map positions of the SNPs
• Chromosome list of chromosomes the SNPs belong
  to
• Idnames list of subjects used in analysis
• Lambda inflation factor estimate, as computed
  using lower portion (say, 90) of the
  distribution, and standard error of the estimate
• Formula formula/function used to compute
  p-values
• Family family of the link function / nature of
  the test

12
GenABEL Table Graphic Functions

• descriptives.marker() table of marker info.
• descriptives.trait() table of trait info.
• descriptives.scan() table of scan results
• plot.scan.gwaa() plot of scan results
• plot.check.marker() plot of marker data (QC
  etc.)

13
GenABELComputer Efficiency
2000 subjects x 500K chip Memory 3.2 G Loading
time 4 Min. SNP summary 1 Min. Call ccfast
0.5 Min. Call qtscore 2 Min. Total lt 10
Min. Permutation test N10,000 73 120 hrs, 35
days
Intel Xeon 2.8GHz processor,SuSE Linux 9.2, R
2.4.1
14
SNPassocAn R package to perform whole genome
association studies, Juan R. González 1, et al.
Bioinformatics, 2007 23(5)654-655

• SNPassoc SNPs-based whole genome association
  studies
• This package carries out most common analysis
  when performing whole genome association studies.
  These analyses include descriptive statistics and
  exploratory analysis of missing values,
  calculation of Hardy-Weinberg equilibrium,
  analysis of association based on generalized
  linear models (either for quantitative or binary
  traits), and analysis of multiple SNPs (haplotype
  and epistasis analysis). Permutation test and
  related tests (sum statistic and truncated
  product) are also implemented.
• Version1.4-9
• DependsR ( 2.4.0), haplo.stats, survival,
  mvtnorm
• Date2007-Oct-16
• AuthorJuan R González, Lluís Armengol, Elisabet
  Guinó, Xavier Solé, and Víctor MorenoMaintainerJu
  an R González ltjrgonzalez at imim.esgt
• LicenseGPL version 2 or newerURLhttp//www.r-pro
  ject.org and http//davinci.crg.es/estivill_lab/sn
  passoc
• In viewsGenetics
• CRAN checksSNPassoc results

15
SNPassoc Data Summary

• setupSNP(datasnp-pheno.table, infomap.table,
• colSNPs, sep "/", ...)
• summary()
• allele frequencies
• percentage of missing values
• HWE test

16
SNPassoc Association Tests

• WGassociation(yx1x2, data, model
  (codominant, dominant, recessive, overdominant,
  log-additive or all),quantitative , level
  0.95)
• scanWGassociation() only p values
• association() only for selected snps, can do
  stratified, GxE interaction analyses
• Results
• Summary a summary table by genes/chromosomes
• Wgstats detailed output(case-control numbers,
  percentages, odds ratios/ mean differences, 95
  confidence intervals, P-value for the likelihood
  ratio test of association, and AIC, etc.)
• Pvalues a table of p-values for each genetic
  model for each SNP
• Plot p values in the -log scale for
  plot.Wgassociation()
• Labels returns the names of the SNPs analyzed

17
SNPassoc Multiple-SNP Analysis

• SNPSNP Interaction
• interactionPval()
• epistasis analysis between all pairs of SNPs (and
  covariates).
• Haplotype Analysis
• haplo.glm() using the R package haplo.stats
• association analysis of haplotypes with a
  response via GLM
• haplo.interaction()
• interactions between haplotypes (and covariates)

18
SNPassoc Computer Efficiency

• 1000 subjects X 3000 SNPs
• 5 min. import data
• 40 min. setupSNP()
• 30 min. scanWGassociation() only p values
  (including permutation test)
• Memory usage 750 MB