Polygenic scores in diabetic individuals across UKBB

Hi all,

I am calculating PRS for a binary trait (T2D) across UKBB individuals. I am using external GWAS from PRS catalog in order to acquire weights for the base file. I have noticed the weights are given in the form of 'Beta'. I have used beta values to calculate PRS in the target sample using PRSice 2 tool. I have noticed the PRS for every individual is closer to zero which is confusing because it contradicts PRS values in T2D individuals from other studies which are higher.

In the PRSice.best file, the maximum value I found is 0.001 which I am not sure makes any sense or not, or probably I am missing something important during analysis. I still have to incorporate covariates which I will be doing later.

Many studies suggest binary traits should have weights in terms of OR. So if simply convert betas in OR (OR=exp(beta)) in my base file and recalculate the PRS from OR. Is it possible to get greater PRS values. or are these PRS are valid and need to be standarized in some way?

Looking forward for the valuable thoughts and suggestions

Thanks

prsice2 ukbb r

• 1,527 views • link

updated 3.7 years ago by Ram

• written 4.7 years ago by m.shoaib •

1 answer

That's because the default is --score avg, which divide the score by the number of alleles included in the PRS, which gives you a smaller PRS score. This has an advantage of accounting for missingness. If you want a large PRS value, then use --score sum though that brings you no benefit

• 0 views • link

written 4.7 years ago by Sam

Ok, if you are talking about genotype missingness I already performed QC steps on my target data (--maf 0.01 --hwe 1e-6 --geno 0.01). Is it ok, if I will use --score sum as I already performed QC steps prior using PRSice 2? Another point, in PRSice bar plot, the best P-value threshold R2 value is 6e-04, is it suggesting model prediction accuracy is low as R2 value is low?

• 0 views • link

written 4.7 years ago by m.shoaib •

Unless you have 0 missingness, my statement would still apply.

Yes, it seems like your model has no power. Maybe worth checking the amount of snp left after matching your GWAS with your target.

Also worth checking the heritability of the phenotype

• 0 views • link

written 4.7 years ago by Sam