Problem with the plink output file for adjusted Bonferroni test

Question

[debitboro]

Hi,

Can anybody help me about the odd output file yielded by the following command:

plink --noweb --allow-no-sex --bfile input_file --assoc --adjust --out output_file

In fact, I'd like to perform a simple Bonferroni basic association test, and I obtains the following output file (extract for first ten rows):

 CHR         SNP      UNADJ         GC       BONF       HOLM   SIDAK_SS   SIDAK_SD     FDR_BH     FDR_BY
   1   rs7540001        INF        INF        INF        INF        INF        INF        INF        INF 
   1   rs2743476        INF        INF        INF        INF        INF        INF        INF        INF 
   1   rs9435653        INF        INF        INF        INF        INF        INF        INF        INF 
   1  rs11810307        INF        INF        INF        INF        INF        INF        INF        INF 
   1  rs12410470        INF        INF        INF        INF        INF        INF        INF        INF 
   1   rs2873038        INF        INF        INF        INF        INF        INF        INF        INF 
   1   rs1553451        INF        INF        INF        INF        INF        INF        INF        INF 
   1   rs2580567        INF        INF        INF        INF        INF        INF        INF        INF 
   1  rs11578605        INF        INF        INF        INF        INF        INF        INF        INF 
   ....
   ....

This file contains (528000 rows (SNPs)), in particular: 328 rows with BONF = "INF", and 1300 others rows with BONF < 0.05, the remaining SNPs present a higher BONF-value than 0.05. My question is: 1300+328 = 1628 (huge) is really the number of SNPs declared significants (<0.05)? Is it normal to get this kind of file ?

Thanks for you

[aydzhouyuan]

Hi

I also have the same problem, did you solve that now?

Answer 1

[chrchang523]

INF indicates an arithmetic overflow, which may not necessarily correspond to a significant SNP. (I'm curious whether PLINK 1.9 gives the same result, and whether it sorts them to the top of the file. This seems like a PLINK 1.07 bug to me.)

As for the other 1300, it is likely that you need to correct for population stratification: one way to get so many significant SNPs is if your phenotype is associated with membership in a particular subpopulation, and all of the SNPs in question are associated with that as well. Other forms of relatedness may also need to be accounted for. This is an active research area; look into "mixed-model association" with principal component covariates. (PLINK does not currently support this, but there are several other good software packages which do, and are capable of reading PLINK-format files.)