Association Modeling With iPlant. Goals of this Section Familiarize with the basic concepts of quantitative genetics: – Traits, phenotypes, genotypes.

Association Modeling With iPlant

Goals of this Section

• Familiarize with the basic concepts of quantitative genetics:– Traits, phenotypes, genotypes

• Understand the basics of trait mapping• Understand the conceptual foundations of

association studies• Lear how to perform a genome wide association

study in the iPlant Discovery Environment– Obtain genotypes– Run a Mixed Linear Model

PhenotypeObservable (measurable) trait (character) of an organism

Trait: eye color

Phenotype: wild type (red), white eyed, orange eyed

http://www.unc.edu/depts/our/hhmi/hhmi-ft_learning_modules/fruitflymodule/phenotypes.html

Qualitative Traits

Campbell, 8e

Controlled by One Locus

Donahue, R. P., et al., Probable assignment of the Duffy blood group locus to chromosome 1 in man, Proceedings of the National Academy of Sciences 61, 949-955 (1968).

Co-segregation in Pedigree

Quantitative Trait

Carlos Harjes

Trait Varies on a Continuous ScaleFr

eque

ncy

Trait Value

Quantitative Traits

• Probably caused by multiple loci– Interaction effects– Environment

If the mean trait value for individuals with marker state MM is different from the mean

trait value of individuals with marker state mm (i.e. the marker is associated with the

phenotype), then the marker is linked to a quantitative trait locus.

Mar

kers

Individuals

Trait value

Marker #3 Mean Trait Value

Present 99 ± 5

Absent 118 ± 8

Marker #6 Mean Trait Value

Present 110 ± 10

Absent 115 ± 13

Quantitative Genetics

Exploring the Genetic Architecture* Underlying Quantitative Traits

*Genetic Architecture• How many loci?• Which location?• How strong?

Tools for Statistical Genetics in the DETool Purpose

Genotype by Sequencing Workflow Automatic pipeline for extracting SNPs from GBS data (with genome from user or from iPlant database)

UNEAK pipeline Automatic pipeline for extracting SNPs from GBS data without reference genomes

MLM workflow Automatic workflow for fitting Mixed Linear Model

GLM workflow Automatic workflow for fitting General Linear Model

QTLC workflow Automatic workflow for composite interval mapping

QTL simulation workflow Automatic workflow for simulating trait data with given linkage map

PLINK PLINK implementation of various association models

Zmapqtl Interval mapping and composite interval mapping with the options to perform a permutation test

LRmapqtl Linear regression modeling

SRmapqtl Stepwise regression modeling

AntEpiSeeker Epistatic interaction modeling

Random Jungle Random Forest implementation for GWAS

FaST-LMM Factored Spectrally Transformed Linear Mixed Modeling

Qxpak Versatile mixed modeling

gluH2P Convert Hapmap format to Ped format

LD Linkage Disequilibrium plot

Structure Estimation of population structure

PGDSpider Data conversion tool

GLMstrucutre GLM with population structure as fixed effect

A Model for Quantitative Traits

P = G + E + GG + GEP=PhenotypeG=GenotypeE=EnvironmentGG=Interaction between genotypesGE=Interaction between genotype and environment

P = G + e

Phenotype

Genotype Environment

A Statistical Model for QTLs

P=G + e

yij trait value in individual j with genotype iβ0 population average of trait valueβ1 effect of marker i on trait valuexi marker genotype iεij error term

General Linear Model (in matrix notation): Y=Xb + e

Note: If errors are not normally distributed, use generalized linear models

http://concord.org/publications/newsletter/2009-spring/genetics

Linkage Mapping (QTL Mapping)

• Designed population– F2– Recombinant inbred (RIL)– Double-Haploid (DH)– Back-cross (B2)

Limitation of Linkage Mapping

• Needs large number of related individuals• Resolution limited (interval contains 100s of

genes)• QTL position and effect are confounded

Association Mapping

• Use random collection of individuals from natural population

• Very dense marker map = very high resolution

Linkage & RecombinationRecombination causes linkage decay

Other factors affecting LD:• Selection (artificial or natural)• Drift• Mutations• Population structure• Demography

Linkage Disequilibrium

Pitfalls: Population Structure

• Difference in allele frequencies between subpopulations

• Due to neutral or adaptive processes

• Can create spurious association

T G T G

No association within groups

• Similar effect due to presence of related individuals (esp. in plants)

• Can be accounted for using the data:– Estimate number of subpopulations– Assign individuals to subpopulation– Estimate kinship

Accounting for Random Effects: Mixed Linear Models

• "Cost" associated with estimating a parameter• We are not interested in the value of the parameter, only the variance• Q-K method (structured association)

y=Xβ+Sα+Qv+Zu+e

Fixed effects:β Vector of fixed effectsα Vector of SNPs effectsv Vector of subpopulation effects

Random effects:u Vector of kinship effectse Residuals

Q Matrix of population association (STRUCTURE)X, S, Z Incidence Matrices

Traits

Markers

Population Structure

Kinship

STRUCTURE

TASSEL

MLM

Obtain Markers

Genome Resequencing Workflow

Genotyping By Sequencing

MLM Pipeline for GWAS

marker

trait

filter

convert

impute

impute

K

GLM

MLM

http://www.maizegenetics.net/statistical-geneticsZhang et al. Nature Genetics. 2010; doi:10.1038/ng.546

Ed Buckler (Cornell University)TASSEL

http://www.maizegenetics.net/tassel/docs/Tassel_User_Guide_3.0.pdf

MLM Input Files

• Hapmap file• Phenotype data• Kinship matrix*• Population structure*

straintraits

Phenotype data

strain3 populations sum to 1

* Kinship matrix & population structure data can be generated using TASSEL or with “MLM Workflow” App in DE

Population structure

Origin

• Hapmap file: – Download (e.g. http://triticeaetoolbox.org/)– Convert from PLINK (.map/.ped) using Tassel 3 Conversion– Impute with NPUTE– Transform to numerical format with NumericalTransform

• Phenotype data• Kinship matrix

– Generate from hapmap marker data with Kinship• Population structure

– Generate using ParallelStructure– Convert to matrix with Structure2Tassel

MLM Output• MLM1.txt

– Marker– “df” degrees of freedom– “F” F distribution for test of marker– “p” p-value– “errordf” df used for denominator of F-test– etc.

• MLM2.txt– Estimated effect for each allele for each marker

• MLM3.txt– The compression results shows the likelihood, genetic variance, and error variance for

each compression level tested during the optimization process.

See TASSEL manual for details:http://www.maizegenetics.net/tassel/docs/Tassel_User_Guide_3.0.pdf

THANKS!