How To Combine P-Values Obtained With Different Approaches?

Question

[J.F.Jiang]

Hi ALL,

I am stacked by a problem of how to combine different p values.

I just got 3 different pvalues using different method, for example chisq test, logistic regression, how can I obtain the combined p? one possible way reported was to convert the p value to z value first and then calculate the final p, is there any other possible method?

Another similar question is that suppose I got 2 GWAS datasets of the same population but different genotype platform though using imputation process, the p value for the same SNP would be different, so how can i combine this kind of p? I know meta can do this kind of thing, but since there is only two datasets, which will lead to impossible for meta analysis

So if anyone knows any method that can fix this kind of problem, could you plz tell me the reference?

Thanks a lot!

Added: not on the same datasets, for example, using a gene expression dataset, I can obtain the p value for a specific gene, meanwhile, using a gwas dataset, I can assign the p-value of most significant SNP to this gene. My question is how to combine these kind of p-values to obtain the updated p-value for this gene?

Answer 1

[Manu Prestat]

You don't want to merge p-values sets obtained by testing different statistical models. If you want to see if 3 different statistics give you similar answers, you can check if they agree by drawing a Venn diagram. I can not tell you more about multi-platform SNP analysis though...

[J.F.Jiang]

not on the same datasets, for example, using a gene expression dataset, I can obtain the p value for a specific gene, meanwhile, using a gwas dataset, I can assign the p-value of most significant SNP to this gene. My question is how to combine these kind of p-values to obtain the updated p-value for this gene?

Answer 2

[Andreas]

EDIT: please see Sean's comment below. Keeping my answer here only for the sake of completeness/discussion.

You might want to look at Fisher's method.

Andreas

[Sean Davis]

Note that Fisher's method is an approach for combining p-values, but it is usually applied in the setting of different dataset, not for multiple p-values from the different tests on the same datasets, as Manu points out.

[J.F.Jiang]

not on the same datasets, for example, using a gene expression dataset, I can obtain the p value for a specific gene, meanwhile, using a gwas dataset, I can assign the p-value of most significant SNP to this gene. My question is how to combine these kind of p-values to obtain the updated p-value for this gene?

Answer 3

[Maxime Lamontagne]

For your GWAS problem, use the Fisher's method for meta-analysis: http://en.wikipedia.org/wiki/Fisher%27s_method

Answer 4

[lyz10302012]

Did you find your answer? I don't think Fisher's method works for your problem. Your problem tends to accept the max P value, but Fisher's method tends to accept min P Value. Any comment?