PRSice output is a bit strange

[Marya]

Hi all,

I used PRSice (2.1.6) with default parameters for calculating the polygenetic risk score for a complex disease. But, I’m not sure about the results. Based on the “.prsice file”, there is one region with R2 of 0.01 and p-value of 4.9e-20. Also, the maximum PRS in the “.best file” is 0.06, and the score is almost similar between case and control individuals, unlike what I expected. Could you please kindly let me know what is wrong here? Also please let me know what is the range of acceptable R2?

 

Thank you in advance

[Sam Choi]

First, maybe try using 2.2.0.

The R2 seems accetable, as for the PRS being similar between case and control, it is difficult to determine if there's any problem w.r.t without looking at your data. Based on information you've provided, it seems like everything is alright

[Marya]

Hi Sam,

Thanks. So, the resulting risk score is useless as it similar between case and control, yes? I think when PRSice reports "there is one region with R2 of 0.01 and p-value of 4.9e-20", I should have an appropriate output (score that can distinguish case from control), so it isn't right, yes? Could you please let me know if there is any way to improve the work?

[Sam Choi]

The phenotypic variance explained by the PRS is 0.01, so in some sense, the PRS is providing some information in case control separation. But if you are trying to accurately predict the case control status using only PRS, that'd be very difficult at lease with the current technology.

So it really depends on the aim of your study.

[Marya]

Hi Sam,

Thank you for your response.

No, I don’t want to accurately predict the disease occurrence, just want to separate case from control. Sorry, in my case, if the R2 of 0.01 mean the power of the used SNPs to distinguish the case from control is just 1%? Please kindly let me know if it shows anything else.

Another thing, if the copies of risk alleles (0/1/2) are randomly assigned to each SNP, the expected value of the risk score will be 0.5; however, in my analysis, the maximum PRS was 0.06. Could you please share your idea about it, if it may show that the SNPs (base file) is used to PRS calculation is not really risky in the target?

 

Thanks in advance

[Sam Choi]

You need to also consider the GWAS effect size. If the SNP is not associated with the disease, the GWAS effect size should be close to 0. As a result of that, for a null trait, we'd expect the PRS to be close to 0

What you really want is the AUC instead of the R2. R2 represents the phenotypic variance explained. You can calculate the AUC using the best score output from PRSice.

[Marya]

Hi Sam,

Thank you for your feedback. I also consider effect size of variants in the base file for PRS calculation, all SNPs with OR > 1 and the p-value cutoff of 5e-8 were associated with the disease of interest (polygenic disease). However, I couldn't understand why the maximum PRS calculated by PRSice is 0.06 in my analysis; assuming that if the copies of risk alleles (0/1/2) are randomly assigned to each SNP,  I expected the risk score value is around 0.5.  I will be happy if you kindly share your opinion with me. 

You kindly suggested computing AUC using the best score output; thanks for your suggestion, could you please let me know if this analysis will be useful when R2 is very low (say 0.01)? also, it will be great if you introduce me an appropriate tool to feed calculated PRS and compute the AUC.

Many thanks in advance