Discrepencies of results with plink --genome using the whole or subset plink file

[Nicolas Rosewick]

Hi,

I've a strange behaviour of plink 1.9 using --genome on my file.

I know that I have a pair of duplicate sample in my file i.e. sample_A and sample_B

Using the whole file (3980 samples) I got for the pair sample_A , sample_B the following result

FID1 IID1     FID2     IID2 RT EZ     Z0     Z1     Z2 PI_HAT PHE      DST    PPC  RATIO IBS0  IBS1   IBS2    HOMHOM    HETHET
1126 sample_A 3784 sample_B UN NA 1.0000 0.0000 0.0000 0.0000  -1 0.892661 0.5227 2.0039 6965 52244 249039 1283.0000 2571.0000

PI_HAT of 0.00 ... strange

However when I only restrict my file to the two sample of interest using --keep command I got :

FID1 IID1     FID2     IID2 RT EZ     Z0     Z1     Z2 PI_HAT PHE      DST    PPC  RATIO IBS0  IBS1   IBS2    HOMHOM    HETHET
1126 sample_A 3784 sample_B UN NA 0.2390 0.0000 0.7610 0.7610  -1 0.892661 0.5227 2.0039 6965 52244 249039 1283.0000 2571.0000

here different results and a PI_HAT of 0.7610 .. 

Any idea what is happening here ? Am I missing something obvious ?

Thank you

[Nicolas Rosewick]

When I try with a bigger subset of my samples (including the pair of sample of interest) : I get again a different result :

FID1     IID1 FID2     IID2 RT EZ     Z0      Z1      Z2  PI_HAT PHE       DST     PPC   RATIO    IBS0    IBS1    IBS2    HOMHOM    HETHET

1126 sample_A 3784 sample_B UN NA 0.9075  0.0000  0.0925  0.0925  -1  0.892661  0.5227  2.0039    6965   52244  249039 1283.0000 2571.0000

Pi_HAT of 0.0925 ... 

[Christopher Chang]

From the --genome documentation: "This estimator requires fairly accurate minor allele frequencies to work properly. Use --read-freq if you do not think your immediate dataset's empirical MAFs are representative."

For identification of close relationships, --make-king[-table] is now strongly recommended over --genome, because it doesn't use allele frequencies.