ECON 546 Lecture Notes - Lecture 3: Atrial Fibrillation, Low-Density Lipoprotein, Epigenetics
27 Jun 2018
School
Department
Course
Professor
BIOL 568 – Jamie Engert L3 & 4: 16,18/1/18
L3 & 4: GWAS
Agnostic – let genetics leads us to discover the genes, without any prior biological clue as to how
they function. Going genome wide. Linkage studies also genome wide.
Genotype/phenotype correlations using Linkage or Association
oLinkage used for lower allele frequency, association for higher. Both poor
for weaker effect size.
Association studies (case vs. control):
oCollect a series of affected and unaffected individuals (cases and controls).
oPick a likely locus and a candidate allele that might have caused the
disease
oCompare the allele or genotype frequencies between the two groups
oConsideration and caveat: success with this approach depends on knowledge of the
underlying biology (e.g. how good is your candidate locus?)
oPotential for greater power for common variants from association vs. linkage
oAssociation studies rely on LD rather than linkage and extends over much smaller regions
oRequire genotyping – but the huge number of markers needed to cover the genome only
became technically feasible ~2006.
oSequencing (whole genome or exome) is much more expensive.
oMillions of human SNPs are now listed in public SNP databases e.g. dbSNP
oOnly a proportion of these SNPs need to be genotyped to capture common genetic variation
through LD.
oDNA samples are collected from populations whose characteristics are well defined (usually
clinical patients with a specific disease and controls).
oFollowed by large-scale genotyping and statistical analysis.
oHas enabled the identification of many novel genes relevant to many particular diseases.
Technology: genome-wide arrays (many choices) for genotyping (Monpetit)
Linkage Disequilibrium (LD)
oLD: non-random association of alleles at different loci in a given population
oWhich SNPs to genotype? Relatedness between two polymorphisms near each other on a
chromosome – don’t need to genotype everything. Imputation.
oLD decreases with distance and with time (no. generations) due to recombination.
oTwo SNPs on different chromosomes are always in equilibrium i.e. LD only between SNPs on
same chromosome.
1
find more resources at oneclass.com
find more resources at oneclass.com
BIOL 568 – Jamie Engert L3 & 4: 16,18/1/18
oHaplotype is the pattern of alleles on single chromosome
oPopulation of chromosomes – all 4 possible haplotypes
oLD describes the allelic association between two SNPs. Two popular LD statistics: D’ and r2
Perfect LD:
oEqual allele frequency; allelic association is as strong as possible
o2 haplotypes observed, can predict position of other. No detected recombination between
SNPs; genotype is perfectly correlated D’ = 1, r2 = 1
oCaveat: different ethnic groups
Complete LD:
oUnequal allele frequency; allelic association is as strong as possible
o3 haplotypes observed. No detected recombination between SNPs; genotype is not perfectly
correlated D’ = 1, r2 < 1
Factors affected LD:
oPhysical distance between SNPs
oDiffering recombination rates
oTime
oSample size
Average LD:
oAt same distance there are SNPs at different LD – graph is missing number of generations.
oProbably two SNPs of different ages
Coalescent Model: Hitchhiking
oTightly-linked neutral alleles are along for the ride.
Haplotype blocks to display LD: more LD near the top (SNPs closer together).
Design and Statistical Issues
1. Study design
2. Choice of genetic polylmorphisms (SNPs)
3. Data quality control
4. Statistical analysis
5. Multiple testing
Study design
Parameters determining power:
2
find more resources at oneclass.com
find more resources at oneclass.com
BIOL 568 – Jamie Engert L3 & 4: 16,18/1/18
a) Sample size – meta-analysis ~200,000
b) Minor allele frequencies: rare variation, fewer people will have it
c) Genetic model (usually additive)
d) LD between marker and disease variant
e) Number of SNPs ( adjusted for multiple testing)
f) Uniformity of SNP coverage – some areas with little SNPs
g) Analysis strategy
h) Staged studies: % of individuals in Stage 1 and % of SNPs carried over to stage 2
Stage Design:
oReducing costs:
Split sample designs:
Two stage: GWAS replication of significant SNPs
Shared pool of control subjects: pooled controls – can be controls for any disease
oMaximising information:
Choosing most extreme (genetically loaded) subjects
Choosing most accurately and comprehensively phenotyped subjects
Two stage: Montreal T2D GWAS
oStage 1: 1000, 400 000 SNPs
oStage 2: 5000, 59 SNPs (fast-track)
oStage 3: replication in additional populations
WTCCC:
o7 diseases, 2000 cases for each disease and 3000 common controls. Database of controls.
oThere is potential for misclassification bias as phenotyping is not available for the shared
control group.
oAssumed that some controls will become cases… should result in loss of power, should not
lead to false positives.
oIt was estimated that if 5% of controls meet the definition of cases, loss of power is approx the
same as that due to reduction of sample size by 10%
oHowever, hypertension might have had 30% misclassification bias
Choice of SNPs
3
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Agnostic let genetics leads us to discover the genes, without any prior biological clue as to how they function. Genotype/phenotype correlations using linkage or association: linkage used for lower allele frequency, association for higher. Technology: genome-wide arrays (many choices) for genotyping (monpetit) Relatedness between two polymorphisms near each other on a chromosome don"t need to genotype everything. Imputation: ld decreases with distance and with time (no. generations) due to recombination, two snps on different chromosomes are always in equilibrium i. e. ld only between snps on same chromosome. L3 & 4: 16,18/1/18: haplotype is the pattern of alleles on single chromosome, population of chromosomes all 4 possible haplotypes, ld describes the allelic association between two snps. Perfect ld: equal allele frequency; allelic association is as strong as possible, 2 haplotypes observed, can predict position of other. Snps; genotype is perfectly correlated d" = 1, r2 = 1: caveat: different ethnic groups.
