BIN300 W5 Spurious associations

Spurious associations due to

mixture of populations, black cows more milk than red cows, association between color locus and milk production

family structure, some families with high milk production, same families happen to have allele A (from particular sire), association between A-allele and milk, A-allele may be at other chromosome than milk

GWAS is based on the regression equation

In statistics: spurious association =

non causal relationship, pure coincidence, other confounding factor that creates the relationship

spurious assocaition

human disease genetics

avoided by balancing the cases and controls

• within every family with a disease case also pick a control individual

  • doubles genotyping costs (otherwise controls could be taking form earlier study)

livestock spurioius association

few large families, family structure is very strong, important ot account avoid spurous association

Q-Q plot

Quantile-Quantile plot

shows associaon problems

expected value of a statistic (with known distribution) against actual value

e.g. highest expected t-statistic (from t-distribution) agianst highest t-value found

under Ho: straight line with slope 1

GWAS study: plot expected P-value against actual P-values

expected P-value under H0: uniform distrbution between 0-1

plot -log(Pvalue)

to spread Pvalues that are close to - (i.e. significant Pvalues)

minus to make high values significant (instead of low values)

QQ plot

H0: expected points on black line

Blue points/: ok results, most points on black line, highest Pvalues exceed black line, significant results

black pints: spurious association: low Pvalue points show inflated values, dont trust most extreme results

lambda is slope of points, measure of spurious association, lambda=1: no spurious association

definition of spurious association

non-random association between SNP and QTL at other chromosome

more broad definitions: SNP and QTL at same chromosome, but at different positoin, this may be seen as QTL signal that is not at the exact QTL position, associaiotn between SNP and enviornmental effects

ass

assocation betwen SNP and QTL at different chromosome?

SNP and QTL come from same pedigree

• if pedigree relationships strong: can predict QTL and SNP

  • thus SNP can also predict QTL, even when on different chromosome

Transmission-disequilibrium test (TDT) test

correct for family effect (pedigree is same within family)

tests whether effect is same within all families, fam uses a lot of degrees of freedom, relatively low power of the test

polygenic effect (A)

fit effect of background genes for every animal

A coorelation matrix based on pedigree, e.g. fullsibs have 50% of genes in common Aij =0.5, instead allocating i and j to family or not → more finescale distinctions: how many genes have i and j in common

requires variance component analyses for every snp

estimate sigmaa2 and sigmee2 for every SNP

polygenic effect G

use polygeneic effect with correlation matrix G (genomic relationships)

G uses actual SNPs on SNPchip to determine how many alleles in common, accounts for what happens at the genome, e.g. fullsibs may have 45% of genes in common instead of 50% (as assumed from pedigree)

tests whether SNP x explains more than the average SNP on the SNPchip

H0: SNP x explains about the same variance as any other SNP in the genome, instead of H0: SNP x explaisn no variance

Does this solve the spurious assication problem

TDT

G MATRIX

A MATRIX

Q-Qplot

TDT- test: yes

Fitting polygeneic effect with G matrix: yes

Fitting polygeneic effect with A matrix: no, may have more or less alleles in common than indicated by pedigree

test whether spurious associations are solved by Q-Q plot

spurious association in livestock

very important, due to mixture of poulatuions and family structure

can be detected by q-q plot, may be solved by TDT test, fitting polygeneic effect, using pedgree relatinship matrix A and G