I think Widen et al. (2022) uses actual sibling pairs/trios (unless I’m misreading?), but a few other studies use simulated embryos such as Lencz et al. (2021) [1] and Turley et al. (2021) [2].
[1] “Utility of polygenic embryo screening for disease depends on the selection strategy”
[2] “Problems with Using Polygenic Scores to Select Embryos”
If you look at the discussion section they say “Specifically, we validated this index in selection experiments using unrelated individuals and sibling pairs and trios from the UK Biobank.”
The graph of relative risk reduction I placed in the post shows reductions among groups of five. They state elsewhere in the paper that there were 969 trios available in UK Biobank, the source of data for the paper. There is simply no way they could have produced confidence intervals as tight as those shown in the graph from real families of five siblings. I would be surprised if there were more than 100 such families in all of UKBB.
Hence I concluded they must be using pseudosiblings.
My estimates for the expected reduction we’d see moving from pseudosiblings to real siblings are pretty rough, but it’s based on Figure 7 in the paper where they show the relative risk reduction size between siblings and unrelated individuals. There’s a lot of variance in the chart, but it looks like there’s maybe a roughly 20% reduction in RRR when moving from pseudosiblings to siblings.
I think Widen et al. (2022) uses actual sibling pairs/trios (unless I’m misreading?), but a few other studies use simulated embryos such as Lencz et al. (2021) [1] and Turley et al. (2021) [2].
[1] “Utility of polygenic embryo screening for disease depends on the selection strategy”
[2] “Problems with Using Polygenic Scores to Select Embryos”
If you look at the discussion section they say “Specifically, we validated this index in selection experiments using unrelated individuals and sibling pairs and trios from the UK Biobank.”
The graph of relative risk reduction I placed in the post shows reductions among groups of five. They state elsewhere in the paper that there were 969 trios available in UK Biobank, the source of data for the paper. There is simply no way they could have produced confidence intervals as tight as those shown in the graph from real families of five siblings. I would be surprised if there were more than 100 such families in all of UKBB.
Hence I concluded they must be using pseudosiblings.
My estimates for the expected reduction we’d see moving from pseudosiblings to real siblings are pretty rough, but it’s based on Figure 7 in the paper where they show the relative risk reduction size between siblings and unrelated individuals. There’s a lot of variance in the chart, but it looks like there’s maybe a roughly 20% reduction in RRR when moving from pseudosiblings to siblings.