Alleles not consistent across cohorts in plink2

[Parish Bracha]

I am using a .bim file created in plink1.9 to do association analysis on 3 cohorts in plink2 using --glm with a --keep list of IDs, a phenotype file, and a covariate file, then using these outputs to do meta-analysis in METAL.

My issue is that the alleles in the association analysis output are different within the 3 cohorts and the .bim file, which is causing METAL to be unable to work due to expecting the alleles from the first cohort to match the other 2 cohorts. 

When I look at the output from PLINK, there are 3 columns for alleles: REF, ALT, and A1. I can see that one of the cohorts has the same alleles as the .bim when looking at ALT and A1 in its output, while the other 2 cohorts do not. These 2 cohorts have the same alleles as each other in ALT and A1, however. Likewise, when REF and ALT are the same across all 3 cohorts, but this does not match the alleles in the .bim.

Is there a specific reason the alleles are not consistent, or is there a way to force the alleles to remain consistent in plink2 so that METAL can successfully process all 3 cohorts?

Thank you,

Parish

[Christopher Chang]

See the documentation for --glm's 'omit-ref' modifier.

[Parish Bracha]

Thank you for your help. I tried utilizing the omit-ref modifier in the --glm call, and everything seems to be in order in the log file, but the output stays the same as without it. I also tried using --read-freq, and the log reflects the frequency file being successfully read, but the output is the same also. The three cohorts are all subdivided from the same population in the original .bed/.bim/.fam files if that helps any.

[Christopher Chang]

You probably need to fix the REF alleles in the per-cohort filesets; see the --ref-allele and --ref-from-fa flags.  Whenever you use plink 1.9 --make-bed, REF/ALT alleles are usually scrambled.