assessment of genomewide association studies tuan v. nguyen garvan institute of medical research...

Post on 18-Jan-2018

222 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

DESCRIPTION

False positive problem

TRANSCRIPT

Assessment of genomewide association studies

Tuan V. NguyenGarvan Institute of Medical Research

Sydney, Australia

WHICH GENES ?

Gene variants ?

False positive problem

Candidate gene studies: reproducibility problem

600 positive associations between common gene variants and disease reported 1986-2000

J N Hirschhorn et al. Genetics in Medicine 2002

166 were studied 3+ times

6 have been consistently replicated

Introduction to genomewide association studies

Genomewide association studies (GWA)

• Revolution in gene search• Hypothesis-free driven approach• Scan 100,000-500,000 gene variants (SNPs)• Case – control design (>1000 individuals)

Massive number of tests of hypothesis

Recent GWA studies in osteoporosis

• Styrkarsdottir U, et al (2008) Multiple genetic loci for bone mineral density and fractures. N Engl J Med 358:2355-2365.

• van Meurs JB, et al (2008) Large-scale analysis of association between LRP5 and LRP6 variants and osteoporosis. JAMA 299:1277-1290.

• Richards JB, et al (2008) Bone mineral density, osteoporosis, and osteoporotic fractures: a genome-wide association study. Lancet 371:1505-1512.

Some gene variants from GWAGene variant (SNP) Gene or location Trait and P-valuers3736228 11q13 (LRP5) BMD (p = 2.6 × 10-9)

Fracture (p = 0.02)rs3736228 11q13 (LRP5) BMD (p = 6.3 × 10-12)

Fracture (p = 0.002)rs4355801 LRP5rs4988321 11q13 (LRP5) BMD (p = 3.3 × 10-8)

Fracture (p = 0.002)rs2302685 12p12 (LRP6) BMD (p = 0.97)

Fracture (p = 0.95)rs4355801 8q24 (TNFRSF11B) BMD (p = 7.6 × 10-10)rs7524102 1p36 (ZBTB40) BMD (p = 9.2 × 10-19)

Fracture (p = 8.4 × 10-4)rs6696981 1p36 (close to ZBTB40) BMD (p = 1.7 × 10-7)

Fracture (p = 2.4 × 10-4)rs3130340 6p21 () BMD (p = 1.2 × 10-7)

Fracture (p = 0.008)rs9479055 6q25 (1) BMD (p = 6.2 × 10-7)rs4870044 6q25 (1) BMD (p = 1.6 × 10-11)rs1038304 6q25 (1) BMD (p = 4.0 × 10-11)rs6929137 6q25 (1) BMD (p = 2.5 × 10-10)rs1999805 6q25 (1) BMD (p = 2.2 × 10-8)rs6993813 8q24 (OPG) BMD (p = 1.8 × 10-14)

Fracture (p = 0.04)rs6469804 8q24 (OPG) BMD (p = 7.4 × 10-15)

Fracture (p = 0.052)rs9594738 13q14 (RANKL) BMD (p = 2.0 × 10-21)rs9594759 13q14 (RANKL) BMD (p = 1.1 × 10-16)rs11898505 2p16 (SPTBN1) Fracture (p = 1.8 × 10-4)rs3018362 18q21 (RANK) Fracture (p = 0.005)rs2306033 11p11 (LRP4) Fracture (p = 0.007)rs7935346 11p11 (LRP4) Fracture (p = 0.02)

What is the credibility of a GWA finding ?

An observed association with p<0.05 does not necessarily mean

that the association exists.

In 100,000 tests, 5000 positive findings could be false positive

Diagnostic test and association test

Diseased

YES

+ve -ve

NO

+ve -ve

Association

True

+ve -ve

False

Sensitivity

P(+ve | D)

False +ve Power P-value

P(+ve | False)

+ve -ve

What do want we to know?

• Probability of association given observed data (eg posterior probability of association)

or

• Probability of observing data if there is no association (P-value)

Posterior probability of association

• Prior probability of association ()• Power = Pr(significance | association)

Sample size• P-value = Pr(significance | no association)

Effect size

is a function of

What is the prior probability of association for a gene variant ?

Gene search = finding small needles in a VERY large haystack

• Human genome ~3 billion base pairs longBUT: Most are vanishingly rare

• 99.9% identical between any two individuals

• ~90% differences between any two individuals is due to common variants

Hypotheses• Common disease / common variants (CD/CV)

(Reich & Lander 2001, Pritchard et al 2005)

• ~90% differences between any two individuals is due to common variants

Prior probability of association ()

• Common variants in the human population: 10 million (Kruglyak and Nickerson Net Gent 2001)

• No. of genetic variants associated with a common disease ~100 or less (Yang et al, Int J Epidemiol 2005)

Prior probability of association

= 0.000001

A Bayesian interpretation of association

10,000,000 common variants

True association (100) No association (9,999,900)

Significant (95)

Non-significant (5)

Significant (100)

Non-significant (9,999,800)

P(True association given a significant result) = 95 / (95+195) = 48%

Power = 95%; P-value=0.00001

A Bayesian interpretation of association

10,000,000 common variants

True association (100) No association (9,999,900)

Significant (95)

Non-significant (5)

Significant (1)

Non-significant (9,999,800)

P(True association given a significant result) = 95 / (95+1) = 99%

Power = 95%; P-value=0.00000001

P-value and “true” association

P-value in the range of 5% - 0.1% will virtually be false positives even in large scale studies

P-value for a reliable association

P < 5 x 10-5

or P < 5 x 10-8

For 1000 cases and 1000 controls,

p< 10-8 are more likely to be true than false

Some gene variants from GWAGene variant (SNP) Gene or location Trait and P-valuers3736228 11q13 (LRP5) BMD (p = 2.6 × 10-9)

Fracture (p = 0.02)rs3736228 11q13 (LRP5) BMD (p = 6.3 × 10-12)

Fracture (p = 0.002)rs4355801 LRP5rs4988321 11q13 (LRP5) BMD (p = 3.3 × 10-8)

Fracture (p = 0.002)rs4355801 8q24 (TNFRSF11B) BMD (p = 7.6 × 10-10)rs7524102 1p36 (ZBTB40) BMD (p = 9.2 × 10-19)

Fracture (p = 8.4 × 10-4)rs9479055 6q25 (1) BMD (p = 6.2 × 10-7)rs4870044 6q25 (1) BMD (p = 1.6 × 10-11)rs1038304 6q25 (1) BMD (p = 4.0 × 10-11)rs6929137 6q25 (1) BMD (p = 2.5 × 10-10)rs1999805 6q25 (1) BMD (p = 2.2 × 10-8)rs6993813 8q24 (OPG) BMD (p = 1.8 × 10-14)

Fracture (p = 0.04)rs6469804 8q24 (OPG) BMD (p = 7.4 × 10-15)

Fracture (p = 0.052)rs9594738 13q14 (RANKL) BMD (p = 2.0 × 10-21)rs9594759 13q14 (RANKL) BMD (p = 1.1 × 10-16)rs11898505 2p16 (SPTBN1) Fracture (p = 1.8 × 10-4)rs3018362 18q21 (RANK) Fracture (p = 0.005)rs2306033 11p11 (LRP4) Fracture (p = 0.007)rs7935346 11p11 (LRP4) Fracture (p = 0.02)

Number of individuals needed to screen in population and family

Hypothetical gene Fracture risk in

Population Family

Relative risk 5 10

Cumulative risk 40% 80%

Cumulative risk after Rx 20% 40%

Number needed to treat 5 2.5

Frequency of risk “genotype”

0.2% 50%

Number needed to screen 2500 5

How many genes are required for a “good” fracture prognosis ?

Odds ratio

Genotype frequency

Number of genes needed for AUC of

0.70 0.80 0.90 0.95

1.1 5% >400 >400 >400 >400

10% 330 >400 >400 >400

30% 150 >400 >400 >400

1.5 5% 33 100 280 >400

10% 19 50 150 330

30% 9 23 70 160

Assessment of GWA finding

• Genetic components of BMD and fracture

• Finding genes of osteoporosis: a challenge

• Genes can help improve the prognosis of fracture

top related