This is a bad idea on multiple levels; the science is less solid/applicable than realised (though not invalid); we should be seriously cautious about embryo selection for anything; embryo selection needs robust regulation that society can be confident in. I will expand on each >>
The science on multiple genetic scores - called sometimes GRS, sometimes PRS on traits you can measure has a very long academic history (the theory is laid down in the 1930s/1940s) and debated/explored in science publications (and twitter)
For this thread the take homes are that (a) the science is robust (b) there are huge numbers of gotchas and complications. One complication is 'cryptic population structure' which these methods can latch onto. The methods are best set up when there is good randomisation
It ends up being that one can achieve 'pretty good' randomisation of genetics even in observational settings (such as human), but one needs to be careful, in particular with phenotypes with complex regional variation - educational attainment is one such case
Now, to be fair to the people who do this, the main signal is there. The most robust testing of this is when one takes GRS/PRS trained in one setting (say, UK) and tested in another (say, Denmark, Sweden, Korea). The method works.
However this replication between populations doesn't mean that the PRS/GRS doesn't have some 'cryptic population structure' leaking in - so one probably has some inadvertant 'southern UK' signal in there the UK score, nothing to do with biology. It will be subtle, but there
The second major problem is that these methods work only really when the person you want to predict "comes from" the same, nicely random "population". In practice this means outside the large populations (currently, basically only Europeans) the theory goes skewed
Some people think you can 'correct' the skew; others (me included) are not so sure. Certainly it is not clear. So - in practice, an applied scientist can only really say this works on "people with well mixed european ancestry".
And, just to be clear, this "well mixed european ancestry" is not a skin colour, visual thing - quite a few people classified as European ancestry one would caution using this (for example, I've got ~0.5% south asian ancestry, and an 'unusual' Y chromosome for northern Europe)
How to use GRS/PRS is a topic of hot debate in cases where we know the outcome is bad (heart attacks) and the action is broadly safe (giving statins) - responsible scientists are having to work hard to know what to do in practice.
The idea that we can be responsible about educational attainment polygenic risk scores etc, with the action being embryo selection is a *massive* overstep in the science now. BUT - this is not the only reason to be concerned (!) >>
Even if one could get a good, biology only risk score on education attainment - something which I think is possible - we should still be very very wary about using this (or in fact any risk score) in embryo selection
This is because the business of predicting an individual on their genetics does not tell us what the consequence of selecting from a set of possible genotypes on one dimension. The main thing is that one can inadvertantly be selecting other traits inherently on the trait
There are some seriously cautionary tales from animal breeding. One of them is in elite chicken breeding (I always love that phrase!) where polygenic selection led to "better" (growing faster) chickens which ... bullied each other so much that they were *not* good in flocks
The key point here is that the selection for one trait (in this case growth) had an *inherent* selection on another trait, not looked at. Only once they got to a serious flock size from this did they realise this, and then had to go back to select for growth *and* behaviour
An important point here in the human setting is not only is it extremely likely embryo selection for educational attainment has some unintended consequences (how can we know?) but also we don't even know what we collectively want to select for (or allow).
This might sound like I am against embryo selection completely - I am not. In the cases where there is highly penetrant alleles, and where these alleles one can be confident only effect the process (usually; loss of function alleles) - this is a case where one can be confident
In practice this means with current knowledge disease causing loss of function alleles in monogenic diseases. Could I imagine more than 1 gene - yes. Could I imagine a polygenic score - I find it hard - I'd prefer more the few genes involved in the main effects.
The final point is about regulation. You might find this tweet thread convincing, but how do you know I am right and someone else is wrong? There are both technical, science issues, and these intersect with societal issues - what traits can we select for in embryos?
Here I really recommend the UK's Human Fertilization and Embryo Authority - it has a strong legal "box" of things which are illegal, and then an area where a group hears evidence and discusses. The group has non scientists, faith representatives and ethicists.
Its critical that society can feel confident not just in the science but also weighing how we can use the science.
So - these are my three main reasons to be against this development. It is neither appropriate scientifically and shows the need for good regulation. I need to transfer these tweets I know into a blog, which I will do so.
