Strand-Orientation In Meta-Analysis With Plink

Question

[Ryan D]

Using PLINK to do meta-analysis, one uses the form

plink --meta-analysis study1.assoc study2.assoc study3.assoc

But in doing so, if there is a mismatch among alleles from the files you are feeding in (study1.assoc, study2.assoc, etc.), the studies won't be incorporated and the plink.prob file output will give the error "ALLELE_MISMATCH". The problem is that the output going in comes from logistic regression files where the reference allele is the minor allele--which often varies between populations. Is there a way to carry out logistic regression in PLINK while keeping the reference allele constant between studies? Alternately, is there a way to alter the output of the study.assoc files so they contain both alleles so that PLINK can perform meta-analysis. In the present logistic output, they look like so.

study1.assoc
 CHR          SNP         BP   A1       TEST    NMISS         OR       SE      L95      U95         STAT            P
   1   rs17556883   38969651    C        ADD     1150      1.101  0.08288   0.9361    1.295        1.163       0.2449

   study2.assoc
 CHR          SNP         BP   A1       TEST    NMISS         OR       SE      L95      U95         STAT            P
   1   rs17556883   38969651    T        ADD     1150      1.101  0.08288   0.9361    1.295        1.163       0.2449

Above, notice the A1 is C for study 1 and T for study 2. Therein lies the problem.

Thanks for any help.

Rx

Answer 1

[Genotepes]

I think you can use the --reference-allele option.

Checking while writing this I found that adding a file like rs10001 A rs10002 T rs10003 T with a SNP and the ref allele per line (the \n here doesn't work).

and the running the analysis using --reference-allele file_alleles.txt should give the OR for your reference allele.

[Ryan D]

Thanks a lot. It worked. I used the command: plink --bfile all --reference-allele ref.txt --covar pheno_test.txt --covar-name Age,BMI,income --logistic --hide-covar --ci 0.95 --out all

Good reference for the PLINK commands here: http://pngu.mgh.harvard.edu/~purcell/plink/reference.shtml