I'm doing some work with polygenic risk scores and have identified a good GWAS study on which to based it. I am going through files from a colleague and what they seem to have attempted is to combine SNPs from different GWAS studies into one PRS.
My question is: is this ever valid? Obviously the effect sizes are calculated fromm entirely different studies and possibly with different statistical approaches. Is it better/correct to base a PRS from a single GWAS?
Depending on whether the different GWAS studies are on different trait, this can have completely different meaning for the downstream. If the different GWASs are on the same trait, this is what we call meta-analysis (usually achieved using METAL), and is usually preferred as this will boost the power of your PRS (though make sure you don't include samples you use for PRS calculation to the GWAS).
On the other hand, if your GWASs are on different traits, you can do something call multi-trait analysis, usually done using MTAG, GenomicSEM etc to name a few. This will have an impact on the interpretation of your results and you would like to think hard and clear what exactly you are trying to achieve here.
Many thanks for that. Just to clarify - I understand the idea of using METAL to combine GWAS scores - but what if we have two GWAS and some SNP is only included in one study and not the other. Is it only appropriate to include SNPs that were captured in both?
Many thanks for that. Just to clarify - I understand the idea of using METAL to combine GWAS scores - but what if we have two GWAS and some SNP is only included in one study and not the other. Is it only appropriate to include SNPs that were captured in both?
There's a setting in METAL to account for that.