Association Mapping And Fdr (False Discovery Rate) Test?

Question

[Aabha]

While we do marker trait correlation using widely used association mapping method, sometimes non significant p value gives significant q value from the FDR test – why? Can anybody explain about this matter?

Thanks for your response. I am providing details about p value and q value for your kind information.

I got p-value from MLM (Mixed linear method) test using TASSEL software. MLM p-values are nominal tests form individual markers and it is not similar with Bonferroni method. Then I followed Storey's Q-method to calculate FDR (http://www.genomine.org/qvalue/index.html).

To determine Q-value using q value soft and I specify the following optional arguments:

Lambda from 0.0 to 0.90 by: 0.05, Choose pi_0 method :Smoother and I got following output (some part) and I indentify as significant q value whereas original p-value shows non-significant eg. value p-value 0.5385(non) q-value-0.04839474(sig) and p-value 0.415(non) q-value 0.03871215 (sig.)

Some more information about p and q value:

p value q value 4.94E-05 2.02E-05 3.11E-08 1.15E-07 8.43E-05 2.82E-05 1.87E-06 2.30E-06 7.89E-06 4.47E-06 5.83E-06 3.58E-06 0.5385 0.04839474 3.80E-06 3.06E-06 0.000133 3.95E-05 1.99E-05 9.78E-06 0.001 0.000193929 0.415 0.03871215

Waiting for your kind response.

[John]

can you provide details on your outputs ....details on methods you used to calculate q values and p values

[Darren J. Fitzpatrick]

When you say non-significant p-values, do mean non-significant using a correction like Bonferroni but significant using FDR or do you mean non-significant at a nominal threshold like 0.05 but significant using FDR?

[Larry_Parnell]

Yeah, we need some details. Is this a human study? Are the individuals related - family-based study? Which phenotype(s)?

[Aabha]

It is not human study, it's a plant genome. Trait I considered like plant height, weight, etc. I am trying to some QTL related with these trait. It is not family based study.

Answer 1

[Larry_Parnell]

We do a lot of genotype-phenotype association tests in our group - with human populations, some family-based, some not. We never use FDR for correction of multiple testing, but prefer other methods. Moreover, I have never seen such FDR applied to association tests within the many GWAS papers I have read.

Be careful not to use a method just because it is available. I'm not sure that FDR is best approach here.

[David Quigley]

Most likely the question comes from someone doing a quantitative trait analysis (e.g. an eQTL study). FDR methods are a standard approach in this literature, in contrast to GWAS studies.

[Aabha]

Yes you right, I am doing QTL study. FDR approach most appropriate when multiple traits are being compaired or markers are not in extensive LD (Chen and Story 2006)

Answer 2

[kumar.vinod81]

Hi to all, I am also getting problem with FDR adjusted p-values in my plant genome wide association data using GAPIT which at the end provide both p-values and FDR adjusted p-values. Mine is a diploid plant species, genome wide 6000 SNPs and 14 quantitative traits. With some chromosome the p-values are most significant at 0.0001 but FDR is coming very high like 0.999, while for some of the data its coming very good like p-values & FDR adjusted p-value both are .0001. Is high FDR 0.999 mean the SNPs might have very less MAF or they are not in extent of linkage disequilibrium? Please suggest what the reasons of this fluxed FDR rate?

Thanks,

Vinod

[David Quigley]

Please make this a new question and remove this reply; the format of this web site is each reply should be an answer to the posed question, not a new question.