Questions about Polygenic Risk Score (PRS) and PRSice-2

Question

[yiqiangz]

Hi there,

I hear Polygenic Risk Score (PRS) recently by reading the PRSice-2 paper. PRS is simple but interesting. There are few questions on PRS and the tool, could anyone help?

1, In my understanding, PRSice-2 scores new individual/population by summing up the effect sizes of the effect alleles from previous GWAS. However, there are at least two different ways to do GWAS, including single-SNP association test which considers each SNP independently as fixed effect and mixed linear model test which considers all SNPs simultaneously as random effect. The effect size estimated by the two methods differ remarkably. So, which association analysis method is recommended for subsequent PRS calculation?

2, How often is the PRS used for non-additive model GWAS (i.e GWAS under dominant/recessive model)? If I understand correctly, PRS can be used for predicting phenotypes, so is it equivalent to the Genomic BLUP? (i.e. GWAS + PRS = training + prediction = Genomic BLUP).

3, Is it SNP effect size (beta) same as SNP heritability? Could I find the definition of effective allele? Is the effect size of non-effective allele 0 by definition?

4, By playing with the example file in the PRSice zip file using the command below:

./PRSice_linux -b TOY_BASE_GWAS.assoc -t TOY_TARGET_DATA --or --binary-target T

it generates an output file "PRSice.best" in which I believe the phenotype predictions are recorded. It looks like:

FID IID In_Regression PRS CAS_1 CAS_1 Yes -0.00599501328 CAS_2 CAS_2 Yes -0.00631017938 CAS_3 CAS_3 Yes -0.00227495325 CAS_4 CAS_4 Yes -0.00204360007 CAS_5 CAS_5 Yes -0.000830676955

I expect the PRS is the continuous phenotype value or binary outcome of the individual in the target data. But how could I convert the PRS values here into binary outcome?

Thank you.

[yiqiangz]

Thank you. For question 4, how to interpret the PRS score in this case?

[Sam]

You can see it as kind of a genetic burden of each sample

Answer 1

[Sam]

You can read more about the basics of PRS in this paper

1. You can use both, though usually we use summary statistics from the single SNP association test GWAS as they are more abundant
2. They are not exactly the same, BLUP are usually more powerful assuming you have all the raw data. PRS might be more power if you have a super powerful GWAS, but overall, it depends
3. They are not the same. SNP heritability is the heritability contributed by all the SNPs combined. Effect size is the effect of the effective allele when compared to the non-effective allele. E.g. comparing someone with 0 copy of the effective allele (therefore 2 non-effective allele) with someone with 1 copy of effective allele, we would expect the change in phenotype / OR be the effect size
4. PRS is a score which is correlated/ associated with the phenotype. It doesn't need to give you a binary prediction for binary outcomes