Power of linkage analysis
Egmond, 2006
Manuel AR Ferreira
Massachusetts General HospitalHarvard Medical School
Boston
Outline
1. Aim
2. Statistical power
3. Estimate the power of linkage analysis
4. Improve the power of linkage analysis
Analytically
Empirically
1. Aim
1. Know what type-I error and power are
2. Know that you can/should estimate the power of your linkage analysis (analytically or empirically)
3. Be aware that there are MANY factors that increase type-I error and decrease the power of linkage
4. Show how to clean your data to detect presence and minimize the impact of these factors
2. Statistical power
Type-1 error
H0 is true
α
In reality…
Type-2 error
β1 - α
Power
1 - β
H0: Person A is not guilty
H1: Person A is guilty – send him to jail
H1 is true
H0 is true
H1 is true
We d
ecid
e…
xx xxxxx
x xxx
xxxxxx
xxxxx x
xxx
xx xxxxx
x xxx
xxxxxx
xxxxx x
xxx
H0: There is NO linkage between a marker and a trait
H1: There is linkage between a marker and a trait
Linkage test statistic has different distributions under H0 and H1
xx
Where should I set the threshold to determine significance?
x
Threshold Power (1 – β)
Type-1 error (α)
To low High High
I decide H1 is true (Linkage)I decide H0 is true
Where should I set the threshold to determine significance?
x
Threshold Power (1 – β)
Type-1 error (α)
To low High High
To high
Low Low
I decide H1 is true
I decide H0 is true
How do I maximise Power while minimising Type-1 error rate?
x
I decide H1 is true
I decide H0 is true
Power (1 – β)
Type-1 error (α)
1. Set a high threshold for significance (i.e. results in low α [e.g. 0.05-0.00002])
2. Try to shift the distribution of the linkage test statistic when H1 is true as far as possible from the distribution when H0 is
true.
Non-centrality parameter
H0 H1
NCP
Mean (μ)
Variance (σ2)
Central Χ2
df
2*(df)
Non-central Χ2
df + NCP
2*(df) + 4*NCP
These distributions ARE NOT chi-sq with 1df!! Just for illustration.. Switch to R to see what they really look like..
…R
H0 H1
NCP
Small NCP Big overlap between H0 and H1 distributions
Lower power
Large NCP Small overlap between H0 and H1 distributions
Greater power
Short practical on GPCGenetic Power Calculator is an online resource for carrying out basic power calculations.
For our 1st example we will use the probability function calculator to play with power
http://pngu.mgh.harvard.edu/~purcell/gpc/
1. Go to: ‘http://pngu.mgh.harvard.edu/~purcell/gpc/’Click the ‘Probability Function Calculator’ tab.
2. We’ll focus on the first 3 input lines. These refer to the chi-sq distribution that we’re interested in right now.
Using the Probability Function Calculator of the GPC
NCP
Degrees of freedom
of your test. E.g. 1df for univariate linkage (ignoring for now that it’s a mixture distribution)
1. Let’s start with a simple exercise.
Determine the critical value (X) of a chi-square distribution with 1 df and NCP = 0, such that P(X>x) = 0.05.
Exercises
df = 1
NCP = 0
P(X>x) = 0.05
X = ?
Determine the P(X>x) for a chi-square distribution with 1 df and NCP = 0 and X = 3.84.
df = 1
NCP = 0
P(X>x) = ?
X = 3.84
2. Find the power when the NCP of the test is 5, degrees of freedom=1, and the critical X is 3.84.
Exercises
df = 1
NCP = 5
P(X>x) = ?
X = 3.84
What if the NCP = 10?
df = 1
NCP = 10
P(X>x) = ?
X = 3.84
NCP
3.84
NCP = 5
NCP = 10
3.84
3. Find the required NCP to obtain a power of 0.8, for degrees of freedom=1 and critical X = 3.84.
Exercises
df = 1
NCP = ?
P(X>x) = 0.8
X = 3.84
What if the X = 13.8?
df = 1
NCP = ?
P(X>x) = 0.8
X = 13.8
NCP
3.84
NCP = ? = 0.8
NCP
13.8
NCP = ? = 0.8
2. Estimate power linkage
analysis
Why is it important to estimate power?
To determine whether the study you’re designing/analysing can in fact localise the QTL you’re looking for.
You’ll need to do it for most grant applications.
When and how should I estimate power?
Study design stage
Analysis stage
How?
Theoretically, empirically
Empirically
When?
Theoretical power estimation
NCP determines the power to detect linkage
NCP = μ(H1 is true) - df
H0 H1
NCP
If we can predict what the NCP of the test will be, we can estimate the power of the test
4. Marker informativeness (i.e. Var(π) and Var(z))
Theoretical power estimation
Variance Components linkage analysis (and some HE extensions)
zCovVVzVarVVarVr
rssNCP DADA ,ˆˆ
1
1
2
1 2222
2
1. The number of sibs in the sibship (s)
2. Residual sib correlation (r)
3. Squared variance due to the additive QTL component
(VA)
5. Squared variance due to the dominance QTL
component (VD).
^
Sham et al. 2000 AJHG 66: 1616
Another short practical on GPC
The idea is to see how genetic parameters and the study design influence the NCP – and so the power – of linkage analysis
1. Go to: ‘http://pngu.mgh.harvard.edu/~purcell/gpc/’Click the ‘VC QTL linkage for sibships’ tab.
Using the ‘VC QTL linkage for sibships’ of the GPC
1. Let’s estimate the power of linkage for the following parameters:
Exercises
QTL additive variance: 0.2
QTL dominance variance: 0
Residual shared variance: 0.4
Residual nonshared variance: 0.4
Recombination fraction: 0
Sample Size: 200
Sibship Size: 2
User-defined type I error rate: 0.05
User-defined power: determine N : 0.8
Power = 0.36 (alpha = 0.05)Sample size for 80% power = 681 families
2. We can now assess the impact of varying the QTL heritability
Exercises
QTL additive variance: 0.4
QTL dominance variance: 0
Residual shared variance: 0.4
Residual nonshared variance: 0.4
Recombination fraction: 0
Sample Size: 200
Sibship Size: 2
User-defined type I error rate: 0.05
User-defined power: determine N : 0.8
Power = 0.73 (alpha = 0.05)Sample size for 80% power = 237 families
3. … the residual shared variance
Exercises
QTL additive variance: 0.2
QTL dominance variance: 0
Residual shared variance: 0.2
Residual nonshared variance: 0.6
Recombination fraction: 0
Sample Size: 200
Sibship Size: 2
User-defined type I error rate: 0.05
User-defined power: determine N : 0.8
Power = 0.26 (alpha = 0.05)Sample size for 80% power = 1161 families
4. … the sample size
Exercises
QTL additive variance: 0.2
QTL dominance variance: 0
Residual shared variance: 0.4
Residual nonshared variance: 0.2
Recombination fraction: 0
Sample Size: 400
Sibship Size: 2
User-defined type I error rate: 0.05
User-defined power: determine N : 0.8
Power = 0.94 (alpha = 0.05)Sample size for 80% power = 294 families
4. … the sibship size
Exercises
QTL additive variance: 0.2
QTL dominance variance: 0
Residual shared variance: 0.4
Residual nonshared variance: 0.2
Recombination fraction: 0
Sample Size: 200
Sibship Size: 3
User-defined type I error rate: 0.05
User-defined power: determine N : 0.8
Power = 0.99 (alpha = 0.05)Sample size for 80% power = 78 families
Theoretical power estimation
Advantages: Fast, GPC
Disadvantages: Approximation, may not fit well
individual study designs, particularly if one needs to
consider more complex pedigrees, missing data,
ascertainment strategies, etc…
Empirical power estimation
Mx: simulate covariance matrices for 3 groups (IBD 0, 1 and 2 pairs) according to an FQE model (i.e. with VQ > 0) and then fit the wrong model (FE). The resulting test statistic (minus 1df) corresponds to the NCP of the test.
See powerFEQ.mx script.
Still has many of the disadvantages of the theoretical approach, but is a useful framework for general power estimations.
Simulate data: generate a dataset with a simulated phenotype and a marker that explains a proportion of the phenotypic variance. Test the marker for linkage with the phenotype. Repeat this N times. For a given α, Power = proportion of replicates with a P-value < α (e.g. < 0.05).
3. Improve power of linkage
analysis
Factors that influence type-1 error/power
linkage1. Selective sampling
2. Sample size
QTL heritability
Disease prevalence
Residual correlation
Sibship size
3. Deviations in trait distribution
4. Outliers
5. Pedigree errors
6. Genotyping errors
7. Marker informativeness
8. Marker density
9. Genetic map
Pedigree errors
Definition. When the self-reported familial relationship for a given pair of individuals differs from the real relationship (determined from genotyping data). Similar for gender mix-ups.
Impact on linkage analysis. Increase type-1 error rate (can also decrease power)
Detection. Can be detected using genome-wide patterns of allele sharing. Some errors are easy to detect. Software: GRR.
Correction. If problem cannot be resolved, delete problematic individuals (family)
Boehnke and Cox (1997), AJHG 61:423-429; Broman and Weber (1998), AJHG 63:1563-4; McPeek and Sun (2000), AJHG 66:1076-94; Epstein et al. (2000), AJHG 67:1219-31.
Pedigree errors *Impact on
linkage*
• CSGA (1997) A genome-wide search for asthma susceptibility loci in ethnically diverse populations. Nat Genet 15:389-92
• ~15 families with wrong relationships
• No significant evidence for linkage
• Error checking is essential!
http://www.sph.umich.edu/csg/abecasisGRR
Pedigree errors
*Detection/Correction*
Practical
Aim:
Identify pedigree errors with GRR
1. Go to: ‘Egmondserver\share\Programs’Copy entire ‘GRR’ folder into your desktop.
2. Go into the ‘GRR’ folder in your desktop, and run the GRR.exe file.
3. Press the ‘Load’ button, and navigate into the same ‘GRR’ folder on the desktop. Select the file ‘sample.ped’ and press ‘Open’. Note that all sibpairs in ‘sample.ped’ were reported to be fullsibs or half-sibs.
I’ll identify one error. Can you identify the other two?
Genotyping errors
Definition. When the observed genotype at a given locus does not match the true genotype at that locus.
Unavoidable (assay quality, genotyping platform); becoming much lower with most recent genotyping technologies (chip arrays).
Impact on linkage analysis. Can substantially decrease power: e.g. 1% genotyping error can result in ~10-50% loss of power for linkage. Can also increase type-1 error rate.
Detection. Look at: assay failure rate (e.g. 20%), number of Mendelian errors, number of genotypes that imply unlikely recombination events. Can be hard to detect (SNPs)!
Correction. (1) Re-type problematic markers/individuals; (2) Remove the problematic genotypes; (3) leave errors in, but model them appropriately.
Genotyping errors *Impact on
linkage*
-4
-3
-2
-1
0
1
2
3
4
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Ave
rag
e L
OD
Successive lines for 0, 0.5, 1, 2 and 5% error.
Genotyping errors
*Detection/Correction*Detection
Correction
(1) Assay failure rate
(2) Mendelian errors (e.g. SIBPAIR)
(3) Genotypes that imply unlikely recombination events
(1) Re-type problematic markers/individuals
(2) Remove the problematic genotypes
(3) Leave errors in, but model them appropriately.
Genotyping errors
*Detection/Correction*
http://www.sph.umich.edu/csg/abecasisMERLIN
Genotyping errors
*Detection/Correction*
• Genotype errors can change inferences about gene flow– May introduce additional
recombinants
• Likelihood sensitivity analysis– How much impact does each
genotype have on likelihood of overall data
2 2 2 22 1 2 12 2 2 22 1 2 11 2 1 22 2 2 21 1 2 22 1 2 11 1 1 11 2 1 22 1 2 11 2 1 21 1 1 1
MERLIN demoDetect and correct genotyping errors
1. Use input files:
2. I’ll first run the program pedstats to have a look at these files first.
pedstats –d error.dat –p error.ped
You can download these from the MERLIN website error.dat error.ped error.map
There are 20 markers and 1 affection trait for 200 families with 4 individuals each.
3. I’ll then test the trait for linkage with each of the 20 markers using MERLIN, using the ‘--npl’ option for the linkage test.
Note this is done before detecting/correcting genotyping errors!
merlin –d error.dat –p error.ped –m error.map --npl
So before correcting any errors, we get a maximum LOD score
of 1.69 at position 52.680 cM
4. But first we should have looked for genotyping errors. Let’s do that using the ‘--error’ option.
merlin –d error.dat –p error.ped –m error.map --error
MERLIN flagged 7 pairs of
unlikely genotypes.
There’s a very good
chance that these
resulted from
genotyping errors!
5. Let’s delete these unlikely genotypes using pedwipe
pedwipe –d error.dat –p error.ped
pedwipe will read the error.dat and error.ped files, and delete the
genotypes that were stored in file merlin.err produced in the
previous step.
We get 2 new files, wiped.dat and wiped.ped, that do not have those
genotyping errors.
6. We can now check whether we get any improvement in the LOD score after removing those genotyping errors.
merlin –d wiped.dat –p wiped.ped –m error.map --npl
After deleting those 7 pairs of genotyping errors, the LOD score
at position 52.680 cM increases from 1.69 to 2.10, ~24%!
7. But instead of identifying and removing errors, MERLIN will soon allow the user to leave the genotyping errors in and model them appropriately. This is the ‘--fit’ function. Have a look at the MERLIN documentation for more info.
Summmary
1. Statistical power
2. Estimate the power of linkage analysis
3. Improve the power of linkage analysis