1
$\begingroup$

I am using PRSice to compute the PRS over a train set and want to use the coefficient used on the train set to apply it on another set which I will call the test set.

Once I compute the PRS I get a file called analysis.prsice

This is the few first lines


Pheno Set Threshold R2 P Coefficient Standard.Error Num_SNP

- Base 5e-08 0.00550157 0.0659631 -21.0513 11.4492 43

- Base 5.005e-05 0.00396637 0.113759 -61.8545 39.1108 184

- Base 0.00010005 0.00236122 0.220463 -68.9527 56.2742 286

- Base 0.00015005 0.00113035 0.394503 -59.4058 69.7677 382

- Base 0.00020005 0.00072253 0.495129 -55.2605 81.0064 471

- Base 0.00025005 0.000334179 0.642121 -41.5674 89.4432 538

- Base 0.00030005 7.5306e-05 0.825111 -21.391 96.8021 601

- Base 0.00035005 3.95773e-06 0.959527 5.25804 103.612 668

- Base 0.00040005 0.000175506 0.735018 37.723 111.455 743

I thought that I could apply the PRS on another dataset by multiplying the coefficient in the result by the allelic dosage of the corresponding SNPs. However, I am not sure how to know which are the corresponding SNPs of the coefficients.

Also, if there was a command that does that and it would be easier, could you please tell me how to do it?

Thanks

$\endgroup$
2
  • $\begingroup$ How does PRSice work, i.e. what ML is it using? What is the accurarcy of your predictions? Applying a trained data set to predict another data set should be straight forward. You don't need to specify individual coefficients $\endgroup$
    – M__
    Oct 17 '20 at 21:06
  • $\begingroup$ I am not sure what you mean? actually, I am not even sure how you measure PRS accuracy. I know in the C+T method that you add the allelic dosage weighted by the effect size in the GWAS $\endgroup$
    – lalaland
    Oct 19 '20 at 12:37
0
$\begingroup$

The coefficient is the third column from the right, you can tell because some are positive and other negative. When the value is negative the association is a negative correlation. When it is positive (last two) that is more like the result you are seeking. The further away from zero the stronger the correlation (positive or negative)

These are very large values for the coefficients. The SNP is the last column on the right. I would disgard the top two results because the probabilities are too borderline, e.g. if you used Bonferoni's correction.

There isn't enough information on this calculation nor the data set to make any further insights other than you have 7 SNPs showing PRS

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.