Thread by @Graham_Coop

So beyond speculating wildly about eugenic selection of embryos for high IQ, it's worth worrying about what Hsu's company is doing today (genomicprediction.com). They’re providing polygenic scores aimed at embryo selection for the following traits 1/n

From the papers they reference (biorxiv.org/content/10.110…) they’re constructing these polygenic scores based on LASSO. A standard way to choose a sparse set of SNPs in human genetics

They seem to use Lasso without any control for population stratification in their disease predictions. That's really poor if you are going to apply these predictions to choose among potential embryos from same parents, as this structure is absent in such comparisons 3/n

The papers they reference provide no evidence that their methods work in predicting differences among children of the same parents. Do they have any indication that these predictors that they’re giving to parents and doctors actually work? 4/n

Both stratification and indirect effects (science.sciencemag.org/content/359/63…) may seriously confound what seems like good genomic predictions when applied within families. I didnt see any discussion of this in any of the papers. 5/n

I couldn't see any guidance on their website of what ethnicities they claim their embryo selection scores would be suitable for (may have missed something). Yet we know polygenic score predictions often only work well in populations genetically close to their GWAS pop. 6/n

There's no evidence from their published papers that they've demonstrated a lack of pleiotropic consequences of the polygenic scores they're giving parents access to. What else might they be selecting on? ¯\_(ツ)_/¯ 7/n

I apologize if I've missed anything, happy to add this thread if I have. Perhaps they have more info within the company. However, there was very little public info. to judge the claims the company is making. 8/n

Note that none of this gets to the huge ethical issues involved in embryo selection for complex traits. But even the practical issues seemingly arent well addressed here 9/n.

Sorry I initially missed this paper, they have some done sib replication for hypothyroidism and T1D (blue control and orange cases, biobank C & G in graphs below. F a separate sib study). So I was unfair with respect to points 4 & 5 above for those traits sciencedirect.com/science/articl…

Note that this replication doesn't seem great: "for sibling pairs where 1 sibling was a case and another was a control, the case sibling had a higher PRS 75.05% of the time."

This phrasing from their FAQ seems confusing (genomicprediction.com/faqs/). My understanding is people are just starting to start testing polygenic scores for adults in the clinic

Find the wording of this FAQ (left panel, from genomicprediction.com/faqs/) on the ethics of testing embryos potentially confusing to members of the public.
The linked position piece (right panel) states explicitly that it is about monogenic disorders

The ethics position piece explicitly notes that embryo tests for multifactorial diseases were not known at the time of writing. I'm also guessing they're talking about testing small numbers of well-validated loci when talking about multifactorial diseases asrm.org/globalassets/a…