Twin Samples and PRS

When working with twin simples and PGS, what is the preferred approach for accounting for relatedness? Is it sufficient to model family structure using a random effect (e.g., 1 | FAMID), or is it preferable to account for relatedness using the full genetic relationship matrix (GRM)? If the latter is recommended, how is this typically implemented in practice, and which software is commonly used?

Replying to “When working with twin simples and PGS, what is th…”:

At QIMR we use GCTA with a GRM and add the PRS as a quantitative covariate

Replying to “When working with twin simples and PGS, what is th…”:

Or you can include a PRS in an OpenMx model or similar

Replying to “When working with twin simples and PGS, what is th…”:

Thanks Sarah!

Replying to “When working with twin simples and PGS, what is th…”:

If you post to the forum we’ll share some GCTA code

Replying to “When working with twin simples and PGS, what is th…”:

Ok, I Will do it thanks

Replying to “When working with twin simples and PGS, what is th…”:

If you’re interested in incremental-R^2, you don’t need to correct for the relatedness, btw. The incremental-R^2 is not biased, the standard error of the beta is

Replying to “When working with twin simples and PGS, what is th…”:

I’ve used GEE (generalized estimating equations) correcting SEs for correlated observations

Replying to “When working with twin simples and PGS, what is th…”:

(package in R)

Replying to “When working with twin simples and PGS, what is th…”:

yes and that would be after calculating PRS, so in the regression step

Replying to “When working with twin simples and PGS, what is th…”:

but does it allow to use the full GRM?

Replying to “When working with twin simples and PGS, what is th…”:

it uses famid, not a grm

Replying to “When working with twin simples and PGS, what is th…”:

it uses family ID and/or twin ID