Question:IMPUTE2: the p-value of snps in the output gen file is different from the original snp in plink.assoc .

Question

[m338102001]

hi,

i use shapeit to do phasing and then inpute2 to generate the gen file.

but after performing association test to the gen file and finally getting a assoc.dosage file, I found that the p-value of original snp (not the imputed snp) is different from the original one (before phasing and impute).

my original plink file is hg19.

here is my comment line:

shapeit:

./shapeit --input_bed mygwas.chr1.bed mygwas.chr1.bim mygwas.chr1.fam \
    --input-map genetic_map_chr1_combined_b37.txt \
    --output-max mygwas.phased.haps mygwas.phased.sample

impute2:

./impute2 -use_prephased_g \
    -m genetic_map_chr_combined_b37.txt \
    -h ALL_1000G_phase1integrated_v3_chr1_impute.hap \
    -l ALL_1000G_phase1integrated_v3_chr1_impute.legend \
    -known_haps_g mygwas.phased.haps \
    int xxx xxx \
    Ne xxx \
    -o mygwas.gen \
    -phase

./plink --dosage mygwas.gen format=3 dose=1 skip0=1 skip1=1 noheader --fam mygwas.fam --assoc --out mygwas.imputed.

I think the p-value of original SNP in the mygwas.chr1.bed/bim/fam file should be the same with the output mygwas.imputed.assoc.dosage, but the result showed that they are different. and this really confused me.

Could everyone tell me how should i fix this problem? really thanks a lot

[m338102001]

thank you karl stemm.

I used --assoc command in PLINK to do association test. after imputation, I used the following command to perform association test.

./plink --dosage mygwas.gen format=3 dose=1 skip0=1 skip1=1 noheader --fam mygwas.fam --assoc --out mygwas.imputed.

---

eg. many SNPs has p-value of 1e-06, but after imputation, it only get 1e-04. (sorry I dont know how to use URL to upload the picture)

---

is there any possible that shapeit automatically impute missing genotype in my data? if this really a problem, how can I do to prevent it?

[Maxime Lamontagne]

IMPUTE2 can impute missing data. Check if the p-value of SNPs with a call rate of 100% is still the same.

[m338102001]

thank you, now I know the reason of the changing p-value.

so if I don't want impute2 to perform this autoimpute function, how could I do? because I did'nt see any command that could perform this function in the webpage of IMPUTE2.

[Maxime Lamontagne]

You could replace the post-imputation genotypes with the original. Use PLINK to remove (--exclude) the SNP from your post-Imputation file and add your old genotype with --merge.

[m338102001]

thank you Maxime Lamontangne ^o^

Answer 1

[karl.stamm]

See here for a LOD Score plot along chromosome 1.

http://www.plosgenetics.org/article/info%3Adoi%2F10.1371%2Fjournal.pgen.1000260

I tried to link to figure 4, but please scroll to figure 4 or 5 for an example of a likelihood score plotted against genomic location.

Wouldn't you expect the statistical odds on each SNP to shift when you add more markers?

[m338102001]

dear karl.stamm,

thank you for your reply. but, the p-value of SNP is calculated based on the A/T proportion in samples. let say before imputation, I have SNP A in original data, and after imputation I add SNP B,C,D,E ,but the A/T proportion of SNP A should not be changed, isn't it?

besides, may I ask you additional question that: if the p-value of SNP A have been changed, what is the possible reason for that?

[karl.stamm]

Firstly the A/T proportion may have changed, you need to double-check that. Depending on the imputation, and calling algorithms. Although I suppose it is very unlikely to have changed, it is possible. What does really change is the knowledge of genetic structure. I don't know what test is computing your p-value, and it sounds like you don't either. If this was a single independent SNP test, then it is directly related to A/T proportion, and no other markers should matter. The reason I linked the figures of a LOD Score plot, is because most good association tests will make use of local structure and be influenced by nearby markers. You've added information to the system, so of course a test's result should change. Make a plot of the before and the after and please post them for us to see.