The results of Polygenic Risk Scores(PRS)

Question

[ysx0226]

PRSice was used to calculate the PRS in ADNI dataset, but the results are quite abnormal ,see belows:

When I use the ADNI data in PRSice dir,the results R2 were quite large (eg:0.749622);

When I use the example data in PRSice dir,the results R2 were quite small (eg:0.0133696);

PRSice.prsice on ADNI

Set Threshold   R2  P   Coefficient Standard.Error  Num_SNP

Base    0.0001  0.749622    8.43404e-23 56.3465 5.7326  33

Base    0.00015 0.829959    4.56913e-21 80.9499 8.59464 54

PRSice.prsice on example

Set Threshold   R2  P   Coefficient Standard.Error  Num_SNP

Base    0.00025 0.0133696   8.43169e-06 -0.197266   0.0442903   2

Base    0.0003  0.00824473  0.000456434 -0.225204   0.0642503   3

[Kevin Blighe]

I added prsice tag, as I believe the developer of the program watches this tag.

[ysx0226]

what do you mean by "add prsice tag", is this a parameter of the program?

[Sam]

Nope, he mean adding a tag to your post so that I can be notified.

[ysx0226]

Thank you for your advice~

Answer 1

[Sam]

You are working on two different dataset, it is therefore normal that they are going to have a different result (different R2).

However, here, we have a special case where you've obtain an abnormally high R2 for your ADNI dataset. When it comes to PRS analysis, we usually expect a small R2, with a high R2, it is usually because you have use the same dataset to generate the GWAS summary statistic AND use the same set of sample to generate the PRS. You should make sure that an independent dataset is used to generate the GWAS summary statistics.

[ysx0226]

Thank you, according to your comment, my results seem to be over-fitting

I think I am using a dataset to predict itself, that is a mistake