Genetic Risk Score (GRS)/ OR values

Question

[Juliana]

Hi everyone,

Please...

In order to look for for a difference in genetic (GWAS' SNPs obtained previously in my population study) risk contribution for an interested phenotype, I've been using a weighted method, calculating an average GRS per allele from each SNP for each individual.

However I blocked now. I don't know if I should use the OR values from the previously GWAS SNPs, or if I should do a logistic regression test with the risk all count alleles per SNP in all my sample in order to obtain the weight (OR) values.

Any help?

Thank you so much!

[coleman_jonathan]

Don't quite understand the question - do you have two different datasets:

• A GWAS on the phenotype of interest
• An independent sample which you want to look at the polygenic effects of risk?

If that is the case, you should weight by the OR from the original GWAS - PLINK can do this (--score). Alternatively, you could use PRSice software, which does the whole thing in a handy R wrapper.

[Juliana]

I'm sorry if I was not so clear...

And thanks in advance..

Actually part of my current study sample population was included in a previously GWAS in order to identify SNPs associated with the disease risk (my interested phenotype until then). We found 40 GWAS significant SNP associated with the disease risk.

Now, my global aim is to assess if a weighted GRS model that includes an overall discrimination from those 40 SNPs could better predict not only the risk disease but also with associated clinical characteristics (ex. level of activity of the disease).

However, before this I should calculate an average GRS for each individual, based on my known risk count allele per SNP ( that in the end will allow me do a GRS model with the 40 risk SNPs). For that, should I still use the OR values from the previously GWAS?

[Annika Forsingdal]

Hi Juliana,

Yes, you should use the OR's from the previous sample (base dataset). Your target (study sample population) should be independent from your base dataset. As mentioned above you will overestimate the prediction and variance explained if your target sample is also in your base dataset. I suggest you exclude your target sample from the first GWAS (base dataset), rerun to get OR not driven by target sample, and then use those to calculate GRS in your target sample. See https://www.ncbi.nlm.nih.gov/pubmed/25132410 https://www.biorxiv.org/content/10.1101/416545v1.

I hope this is helpful.

Best, Annika

[ysx0226]

Hi everyone,

Is there another weighted method for SNP after GWAS, such as GRS

Answer 1

[coleman_jonathan]

Use the OR from the previous GWAS.

However, you need to exclude all the people who were in that previous GWAS from your sample, or you will over-fit your model (the phenotype of the people who were in the original GWAS will obviously be well predicted by their GRS score across the 40 SNPs, because you would be using data A to predict data A, if you see what I mean).

[Juliana]

Dear Coleman,

Thank you so much!! It was a perfect explanation.

I have been trying to do my best! As you already might be noticed, I am a beginner in this area! =)

With my best...