PRSice 2 results

Question

[m.shoaib]

Hi,

I am using PRSice 2 for analysing UKBB binary phenotype (cases vs controls). This is how my PRSice.summary looks like:

Threshold  5.005e-05
PRS.R2 0.0144168
PVALUE   6.99334e-195
Num_SNP 333

What I understood, that with p-value threshold 5e-5, the model prediction accuarcy is 0.014 which is low but highest in my analysis. With this prediction, is it possible to differentiate the PRS of cases and controls in the validation sample?

The PRS mean for both cases and controls in PRSice target data in the PRSice.best file is almost same and cannot be differentiated. But PRSice still suggest the P-value threshold with possible predictive utility.

I am trying to comprehend my results and thinking to use the best SNPs with their weights to calculate PRS in independent data. Not sure, if this going to lead somewhere or not

Please share your insights and suggestions.

Thanks a lot

Answer 1

[Sam]

Your PRS explains around 1% of your phenotypic liability and this level of prediction is rather common for PRS. Unfortunately, this is also not enough for complete separation of case and control.