Handbook on Analyzing Human Genetic Data

339 Pages·2009·2.41 MB·English

Handbook on Analyzing Human Genetic Data
Shili Lin (cid:129) Hongyu Zhao
Editors
Handbook on Analyzing
Human Genetic Data
Computational Approaches and Software
123
Editors
ProfessorDr.ShiliLin
DepartmentofStatistics
TheOhioStateUniversity
Columbus,Ohio43210
USA
[email protected]
ProfessorDr.HongyuZhao
DepartmentofEpidemiologyandPublicHealth
YaleUniversity
SchoolofMedicine
60CollegeSt.
NewHaven,CT06520-8034
USA
[email protected]
ISBN978-3-540-69263-8 e-ISBN978-3-540-69624-5
DOI10.1007/978-3-540-69264-5
SpringerHeidelberg Dordrecht LondonNewYork
LibraryofCongressControlNumber:2009931713
(cid:2)c Springer-VerlagBerlinHeidelberg2010
Thisworkissubjecttocopyright.Allrightsarereserved,whetherthewholeorpartofthematerialis
concerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation,broadcasting,
reproductiononmicrofilmorinanyotherway,andstorageindatabanks.Duplicationofthispublication
orpartsthereofispermitted onlyundertheprovisionsoftheGermanCopyrightLawofSeptember9,
1965,initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer.Violationsare
liabletoprosecutionundertheGermanCopyrightLaw.
Theuseofgeneraldescriptivenames,registerednames,trademarks,etc.inthispublicationdoesnotimply,
evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevantprotectivelaws
andregulationsandthereforefreeforgeneraluse.
Coverdesign:WMXDesignGmbH,Heidelberg
Printedonacid-freepaper
SpringerispartofSpringerScience+BusinessMedia(www.springer.com)
Preface and Introduction
Thedisciplineofstatisticalgeneticsishighlycomputational.Beitexactcomputa-
tionalmethods,simulationbased,orahybridofthetwo,computationalpackagesare
indispensabletoolsandconstantcompanionsofresearchersinthefield.Thishand-
bookisintendedtoprovidehumangeneticistsandotherbiomedicalresearcherswith
guidance on selections of appropriate computational methods and software pack-
agesfortheirspecificgeneticproblems.Itmayalsobeusedbystudentsandother
learnersasareferenceinconjunctionwithamoretheoreticaland/ormethodologi-
callyorientedtextbook.Thisbooktriestostrikeabalancebetweenmethodological
expositions and practical guidelines for software selections. Wherever possible,
comparisonsamongthecompetingmethodsandsoftwarearemadetohighlightthe
relativeadvantagesanddisadvantagesoftheapproaches,sothatthereadercanmake
informedchoicestobestmatchtheirspecificneeds.
Human genetics has been undergoingan evolution in the past several years as
new knowledge and technologies are transforming the field, leading to numerous
newdiscoveriesofgenesassociatedwithcomplextraitssuchascancer,obesity,and
diabetes. Many recent genome-wide association studies employ the case–control
design,wherethestudy subjectsconsistofunrelatedaffectedindividualsandnor-
mal controls. For each individual, a large number of genetic markers are queried.
Ageneticmarkerreferstoalocationinthehumangenomewherepeoplemaydiffer
inthegeneticmaterialtheycarry.Geneticmarkerscancomeindifferentforms,with
thesinglenucleotidepolymorphisms(SNPs)mostcommonlyusedduetotheirhigh
abundanceinthegenomeandtheavailabilitiesofreliableandaffordabletechnolo-
giestogenotypethem.ForaSNP,twodifferentforms(calledalleles)generallyexist
atasinglenucleotideposition.Becauseeachpersoncarriestwochromosomes,fora
givenSNPwithtwoallelesAanda,therearethreepossiblegenotypesapersoncan
have:AA,Aa,andaa.Inthissetting,ageneticassociationstudyamountstoiden-
tifying markers that are associated with disease status. This can be accomplished
byexaminingwhetherthereisastatisticalassociationbetweenthemarkergenotype
and the disease status. Although this analysis resembles a standard epidemiologi-
calstudywhereeachmarkercanbetreatedasapotentialriskfactor,therearemany
issuesthatareuniquetogeneticsstudiesthatneedtobeaddressed.Forexample,one
majorconcerninthesestudiesissampleheterogeneityintheirgeneticbackground,
andignoringthisissuemayresultinmanyfalsepositivefindingsthathavenothing
v
vi PrefaceandIntroduction
to do with disease etiology. On the other hand, much research has been done to
empiricallycharacterizeandtheoreticallymodelthedistributionsanddependencies
ofgeneticmarkers,andsuchknowledgeisverybeneficialforassociationanalysis.
Infact,athoroughgeneticassociationanalysisisnotpossiblewithoutagoodunder-
standingofthebasicprinciplesinpopulationgenetics,afielddevotedtothestudy
oftheallelefrequencydistributionandchangeundervariousfactorsthatcanimpact
them, including mutations, random sampling, migrations, and natural selections.
ThechapterbyDr.Weir providesanoverviewofthebasic conceptsofpopulation
geneticsandservesasthestartingpointoftheanalysisofhumangeneticsdata.
Althoughcurrentgenotypingplatformscangenotypeuptoonemillionmarkers,
therearemanymoremarkersinthegenomethatarenotqueriedontheseplatforms.
The reason that these typed markers can provide a good coverage of the genome
isthedependenceamongphysicallyclosemarkers,andsuchdependenceis called
linkagedisequilibrium.Forexample,ifoneSNPhasallelesAandaeachwithallele
frequency50%,andanothermarkerwithallelesBandbeachwithfrequency50%.
If the two markers are independent of each other, we would expect that 25% of
chromosomescarrybothAandBinthepopulation,andsimilarlyforallotherthree
possible combinations: Ab, aB, and ab. However, it is often the case that if these
twomarkersareveryclosetoeachotheronthesamechromosome,thetwoalleles
carried on the same chromosome are not independent. In the most extreme case,
there are only two types of chromosomes, those carrying AB and those carrying
ab, a phenomenon called perfect linkage disequilibrium. Haplotypes refer to the
combinationofallelesonthesamechromosome,andthepresenceofsuchmarker
dependency is the key underlying recent successes of genetic association studies
collectingthegenotypesfromonlyasmallfractionofallknownmarkers.Thereare
manystatisticalchallengespresentedintheanalysisofhaplotypes,bothforpopula-
tiongeneticsstudiesandformoreeffectivegeneticassociationstudies.Thesetopics
arediscussedinthechapterbyDrs.ZhangandNiufocusingonpopulationgenetics
and in the chapterby Drs. Epstein and Kwee in the contextof disease association
analysis.
Genetic association studies can be performed on unrelated individuals using
traditional epidemiological designs, for example, case–control design and cohort
design,ordesignsuniquetogeneticstudies,forexample,family-basedassociation
design.Becausesampleheterogeneityingeneticbackgroundisonemajorconcern
in the validity of a genetic association study based on unrelated individuals, var-
ious statistical methods have been proposed to utilize genetic information in the
collected markergenotypesto make appropriateadjustmentsin association analy-
sis. For example, with enough marker information, it is possible to infer genetic
backgroundfor each individualand such inferred backgroundinformation can be
incorporatedin association analysisto make the results less susceptible to sample
heterogeneity.ThisissueisthoroughlystudiedandaddressedinthechapterbyDrs.
ZhuandZhang.
With data from related individuals,genetic association tests may be conducted
inamannerthatisvalid(i.e.,notsubjecttobiasduetosampleheterogeneity)even
withoututilizinggeneticmarkerstoinfergeneticbackground.Thebasicprincipleis
todetectwhetherthereisadeparturefromrandommarkersegregationatacandidate
PrefaceandIntroduction vii
locus. For example, if a study population consists of affected children and their
parentsandamarkerwithtwoallelesAandaisstudiedforitspotentialinvolvement
inthedisease.Ifthemarkerhasnothingtodowithdiseasephenotype,weexpectthat
a parentwho is heterozygousAa wouldhave equalchance to transmitallele A or
atohis/heraffectedoffspring.Ontheotherhand,ifalleleAincreasesdiseaserisk,
we would expectto observe a preferentialtransmission of allele A to the affected
offspring.Thistestingprocedureisrobusttosampleheterogeneityastheinference
is conditional on each parent’s genotype and the only genetic principle tested is
randommarkeralleletransmissionfromparentstooffspring,theMendel’sfirstlaw.
Manystatisticaldevelopmentsalongthisresearchroutearediscussedinthechapter
byDrs.ZhangandZhao.
Both population-basedand family-basedassociation studies examinestatistical
associations between a phenotype and the genotypes at a marker. One implicit
assumptionisthatthesamemarkergenotypewouldexertthesameorsimilareffects
onaphenotype.Whilethisisexpectedtobethecaseformostgeneticmarkersthat
havedirectfunctionalimpact,thisassumptionmaywellbeviolatedformanymark-
ers.Forexample,consideramarkerwithtwoallelesAandastudiedisnotfunctional
butrather is in linkage disequilibriumwith a truly functionalone with two alleles
Dandd.ItispossiblethatAispositivelyassociatedwithDinonepopulation,that
is, someonecarryingA on one chromosomeis also more likely to carry D on the
samechromosome,butAisnegativelyassociatedwithDinanotherpopulation.In
this case, an analysis using samples from these two populationstogether may not
even be able to detect a genetic association. More importantly,when the markers
aresparseandnotexpectedtoprovideagoodcoverageofthegenome,theassoci-
ationanalysisparadigmdiscussedabovewillnotbeeffectiveasalargeproportion
ofthegenomethatlikelyharborsdisease genesmaybemissedduetopoorcover-
age.Thiswasinfactthecaseonlyafewyearsagowhenonlyfewermarkerscould
beusedforgeneticanalysis.Inthisscenario,althoughthemarkerswerenotdense
enoughtocoverthegenomeforassociationanalysis,theyweremorethanadequate
to allowgeneticiststo inferwhethertworelativesinan ascertainedpedigreeshare
asegmentinthegenomefromthesameancestor.Forexample,iftwosiblingshave
thesamemarkergenotypesacrossasetofcloselylinkedmarkersonthesamechro-
mosome,thentheylikelyhaveinheritedthesamegeneticmaterialsfromboththeir
parents. A genetic linkage analysis is to statistically assess whether there is a co-
segregationofgeneticmaterialswithinacandidateregionandthephenotypewithin
a family.For example,this canbedonebystudyingwhetherthereisa correlation
between trait similarities and inheritance similarities at a candidate region among
a set of individuals from the same family. Consider a study enrolling affected sib
pairs. If majority of them share the same genetic materialsfrom their parentsin a
region,thenthisregionislikelyinvolvedindiseaseetiology.Notethatincontrastto
associationanalysisthatis performedacrossallstudysubjects,linkageanalysisis
conductedwithinfamiliesandevidenceisthensummedoveracrossindividualfam-
ilies.Statisticalmethodsforlinkageanalysiscanbeconductedforeitherqualitative
traits(thechapterbyDr.LiandAbecasis)orquantitativetraits(thechapterbyDrs.
Amos,Peng,Xu,andMa).
viii PrefaceandIntroduction
Exactinferenceofinheritancepatternswithinapedigreeistractableeitherfora
small pedigreeor for a few markers, but such inferencebecomescomputationally
prohibitive for large pedigree with many genetic markers. In this case, the exact
probabilitiesmaybeestimatedbyMonteCarlosimulations.InthechapterbyDrs.
Igo, Luo, and Lin, the principles and implementations underlying the simulation
methodsforlinkageanalysisinlargecomplexpedigreesarediscussed.
One central topic in statistical inference is the control of false positive results
soastominimizeanyconsequencesresultingfromfalseleads.Thisissuehasbeen
welladdressedwhenonlyoneorasmallnumberofstatisticalhypothesesaretested.
However, hundreds of thousands of markers are tested for their associations with
disease in a genome-wideassociation study, and false positive controlat the indi-
vidualmarkerlevelswillnotbeadequate.Forexample,ifastudyconsiders500,000
markersandthestatisticalsignificancelevelissetat0.01,wewouldexpecttosee
5,000falsepositiveresultsevenwhenthereisnoassociationbetweendiseasestatus
andanyofthemarkers.Similarissueexistsinthelinkageanalysiscontext,although
nottothesamegreatextentasassociationanalysis.ThechapterbyDrs.Zhangand
Ott presents some recent developments on appropriately controlling overall false
positiveresultsingeneticstudiesatthegenomelevel.
Theidentificationsofdiseasegenescanleadtobiologicalinsightsonpathways
involved in disease etiology, and these findings can also be used to predict an
individual’s disease risk. In the chapter by Drs. Gail and Chatterjee, they discuss
statisticalmethodsthatcanbeusedtomakeuseoffindingsfromgeneticstudiesto
identifyindividualsathigherrisksfordisease.
ThebookconcludeswiththelastchapterbyDrs.Molony,Sieberts,andSchadt,
wheretheydiscussintegratinggeneticsandgenomicsdatatobetterdelineatebiolog-
icalpathwaysunderlyingcomplextraits.Inadditiontodiseasestatusandpossibly
other clinical outcomes, they consider gene expression data that can now be rou-
tinely gathered to measure the expression levels of tens of thousands of genes
simultaneously for each study subject. These gene expression data add another
wholenewdimensionofstatisticalanalysisandareveryinformationrich.Inprin-
ciple, the expression levelof each gene can be thoughtas a quantitative trait, and
linkage/associationanalysiscanbeconductedtoidentitygenesregulatingagene’s
expressionlevel.Therefore,basedonthisperspective,wewouldbeinapositionto
conductgenetic analysis for tens of thousandsof traits. Some of these expression
levels may be associated with disease outcome, and so it is natural to investigate
how a genetic variation affects the expression levels as well as disease outcomes.
Manybiologicalquestionsontheunderlyinggeneticnetworksrelatinggeneticvari-
ations,expressionvariations,andphenotypevariationscanbeposedandanswered
with these data. This chapter discusses topics falling into the domain of systems
biologywherethewholebiologicalsystemisthefocusofastudyandgenome-level
dataofdifferenttypesareneededtodissectthenetworks.
Wehopethatthisbookwillprovideanoverviewofthemostimportantareasin
geneticdataanalysismethods.Wefocusonfundamentalprinciplesand,whenpos-
sible,demonstratetheseprincipleswithrealdataexamples.Despiteourefforts,this
is not an encyclopedia of statistical methods in human genetics, and some topics
PrefaceandIntroduction ix
are not included such as the experimental design of a genetic study, data prepro-
cessing from high-throughputgenotyping platforms, and copy number variations.
Mostimportantly,thisis a veryrapidlydevelopingfield and newtechnologiesare
constantly introduced that demand novel statistical approaches to make the most
useofthedatacollected.Forexample,thestatisticalmethodsdiscussedinthisbook
maynotbethemosteffectiveforinferringinheritancepatternsinapedigreeusing
high density SNP data. On the other hand,the availabilitiesof re-sequencingdata
fromalargenumberofstudysubjectsleadtoanewsetofinformaticsandstatistical
challenges,suchastheincorporationofSNPannotationinformationandthedeal-
ingofraregeneticvariations.Wehopethebasicprinciplesandstatisticalmethods
discussedinthisbookwillmotivatethereaderstodeveloptheirownapproachesif
necessarytoaccelerateourprogressesinmappingdiseasegenes.
Contents
PopulationGenetics................................................................ 1
BruceWeir
HaplotypeStructure............................................................... 25
YuZhangandTianhuaNiu
LinkageAnalysisofQualitativeTraits........................................... 81
MingyaoLiandGonc¸aloR.Abecasis
LinkageAnalysisofQuantitativeTraits.........................................119
ChristopherI.Amos,BoPeng,YajiXu,andJianzhongMa
MarkovChainMonteCarloLinkageAnalysisMethods ......................147
RobertP.IgoandYuqunLuo,ShiliLin
Population-BasedAssociationStudies...........................................171
XiaofengZhuandShuangLinZhang
Family-BasedAssociationStudies................................................191
KuiZhangandHongyuZhao
HaplotypeAssociationAnalysis ..................................................241
MichaelP.EpsteinandLydiaC.Kwee
MultipleComparisons/TestingIssues............................................277
QingrunZhangandJurgOtt
Estimating the Absolute Risk of Disease Associatedwith
IdentifiedMutations ...............................................................289
MitchellH.GailandNilanjanChatterjee
xi
xii Contents
ProcessingLarge-Scale,High-DimensionGeneticandGene
ExpressionData....................................................................307
ClionaMolony,SolveigK.Sieberts,andEricE.Schadt
Index.................................................................................331