Screening Human Embryos for Polygenic Traits Has Limited Utility

By Ehud Karavani,, possibly an important piece:

The increasing proportion of variance in human complex traits explained by polygenic scores, along with progress in preimplantation genetic diagnosis, suggests the possibility of screening embryos for traits such as height or cognitive ability. However, the expected outcomes of embryo screening are unclear, which undermines discussion of associated ethical concerns. Here, we use theory, simulations, and real data to evaluate the potential gain of embryo screening, defined as the difference in trait value between the top-scoring embryo and the average embryo. The gain increases very slowly with the number of embryos but more rapidly with the variance explained by the score. Given current technology, the average gain due to screening would be ≈2.5 cm for height and ≈2.5 IQ points for cognitive ability. These mean values are accompanied by wide prediction intervals, and indeed, in large nuclear families, the majority of children top-scoring for height are not the tallest.

Here is the link, and for the pointer I thank the excellent Kevin Lewis.


Doesn't focusing on the average / expected outcome miss the point? In my view screening is aimed at avoiding extreme negative situations.

Good point.


But more generally, it’s 2019. What was the answer 30 years ago ?

Oh. We couldn’t even ask the question.

Polonium was discovered in 1898. Took a world war mobilization to turn Curie’s scientific discoveries into a nuclear bomb 47 years later.

What will the gain be by 2039? 2139?

For all the Steve Sailers of the world, are you still afraid of immigration if the lowest point of IQ is > 100?

If so, why?

Does Harvard pick randomly among all applicants who get at least a 1000 (50th percentile) on the SAT?

This study is only talking about *polygenetic* traits. Polygenetic traits are the ones that are caused by many genes working together, like height or intelligence. The extreme negative situations are typically caused by a single defective gene. So yeah, screening is typically aimed at avoiding the negative outliers, and this study is confirming that this sort of avoid-the-outlier screening is the only sort that makes sense with modern technology.

But if you are going to screen for bad major mutations, how much more expensive is it to screen for PGS?

And what are the effects over several generations?

yes, it misses the point because slight changes in mean expected outcome can have huge effects on the tail.

even this minor 2.5 pt increase in the mean increases the chance of drawing a baby with a 130+ IQ by over 40%. when you look at it that way, its utility doesn't seem so limited.

But are there big effects in having 40% probability of a baby with 130+ IQ rather than 40% probability of 127+ IQ?

Hive Mind suggests "not really" on a social level, since you get fairly linear scaling over the interval 83->110 in the data, not anything exponential. If you got exponential change with small shifts in IQ at the frontier, would suggest that small changes at the tails in "big brains" lead to big changes in economic activity. But this is not so.

Particularly if you largely extract finance and investment flows (radically changes Singapore's position for'ex) and look at productivity measures rather than raw GDP/capita.

PPP Consumption/hour in some of the major East Asian economies (South Korea, Singapore, Taiwan) is barely above Eastern European levels (and China isn't much above SE Asia in PPP Consumption/hour worked).

Say you need to choose among embryos for implantation for health reasons. Would you bother choosing or let random luck do it's thing?

It's a lot like lesbians looking through the catalog of sperm donors: few pick randomly and most put some effort into picking what they see as the best bet.

"Say you need ...": there being no such need, I would take nature as my teacher and view it as a sign I was not meant to reproduce.

"It's a lot like lesbians ...": I hope not, because the set of men from whom they must choose, if forced to choose from among total strangers, will categorically not include men with a sense of honor and responsibility, should those be inborn traits.

Being a sperm donor is dishonorable? Is that some sort of trad-con thing or what?

Integrity "some sort of trad-con thing"? Okay, let's take this slow. Try - comic-con thing: all the world's a comic-con, and men and women but poor cosplayers upon it. The first rule of integrity club is there are some rules in integrity club. For one, potential children are owed a patrimony that amounts to more than the knowledge that their biological father was willing and successfully able to spill his seed into a cup for money. Everyone begins the game secure of that much; start worldbuilding from there. If you wind up with all the women superpowered ninja warriors and all the men either buffoons or thugs, you've gone wrong and violated canon somewhere. Start over. Welcome to the fandom.

I mean if you can choose then you might as well get that extra 2% of expected intelligence. It doesn't seem like as big of a deal as selecting who the mate is, though. It seems like that would have a far bigger effect on polygenetic traits like intelligence.

First, 2.5cm is 1", that can be a big deal if it you're 5'10" or 5'11", especially to chicks who repeatedly won't date anyone under 6'. Second, 2.5 IQ points - on the margins - could be the dividing line between breaking 100 IQ or not and being able to say or do or concentrate on the right thing at the right time. Third, as the commenters above just mentioned most are selecting out negative traits first. Fourth, as the competitiveness by an ever increasing number of highly ambitious families worldwide increases, these marginal gains could in fact represent something of a marginal revolution in results...every raindrop raises the ocean.

The cost of changing generally only changes when the cost of changing is less than the cost of staying the same. Staying the same in this game for these people is getting quite expensive. If they could take a chance buying an extra inch for their child and an extra 2.5 IQ points for $5 million they're going to do it. That's why it's inevitable this is going to happen.

An extra inch and an extra 2.5 IQ points, whatever that really means, won't make or break a person. Russia's Peter the Great had a personal guard of giant Latvians. Are the descendants of those monster Baltics running Europe now? Are NBA alumni particularly well-noted for success later in life? Maybe in sexual adventure. The criminal world isn't composed of just the stupid. There are many fairly intelligent felons.

2.5 IQ points is the difference between being smarter than 50% of the class and 59% of the class. If lead poisoning resulted in a comparable shift in the opposite direction it would be rightly recognized as a tradgedy. 2.5 IQ points is huge.

If IQ measures smartness, what is IQ and what defines smartness? Is an individual that can read music and compose striking tunes smarter than another that's capable of surviving in an emergency? Does the composer have a higher IQ than the latter? What would be the genetic background for a successful lion tamer? If Al Gore Sr. had picked an embryo for his noted son would Jr. have been a prophet instead of a doofuss?

as any hobbyist playing daily fantasy sports and trying to score in the top 1% each night can tell you, slight increases in mean expected outcome can have an outsized effect on expected tail outcomes.

for example, for population with a mean IQ of 100 with a sd of 15, ~2.3% score above 130. with a mean of 102.5 and sd of 15, ~3.3% score above 130. that seemingly insignificant shift actually increases your chance of having a gifted child by over 40%.

most people do not have an intuition for distributions at all. if they did, they'd never describe something like this as having "limited utility" in a world where almost all innovation is coming from the far right of the distribution.

What does the "mean IQ" have to do with the IQ of any particular person?

It tells you if that person is Albert Einstein or Chief Wiggum.

Charlie Darwin rules!

Biology is ultimately machinery, and even if divinely created, no one looks for angels pushing RNA about. If it's machinery, it's possible that human beings can comprehend it, and do better. But at this point only possible, not certain.

In the meantime, for God's sake people, practice on carrots or potatoes.

I think you mean physiology.

I've never been 100% with the words and whatnot(*).

* - and I steal comedic devices from George Wallace

Hasn't Steve Hsu already pointed out the weaknesses in Shai Carmi's study? "Carmi’s analysis relies primarily on “simulated” data, ours is 100% empirical. We made your writer abundantly aware of the published work validating differentiation of real siblings (not “synthetic genomes”) by polygenic disease status, linking to it in email correspondence:"

a) Everything is of limited utility
b) The utility will increase over time

If we don't figure out how to edit embryos within the next 10-20 years then I think it is likely this could become one of the the largest industries on earth and the repercussions would be impressive.

The very nature of omnigenic/polygenic traits and disorders is that making medical decisions such as screening for an individual (embryo) is likely to have only a tiny effect if any. Behavioral geneticists like Kevin Mitchell have been explaining the genetic architecture of IQ for several years and why genes are often the wrong to try to address traits related to mental processes and behavior. I highly recommend his book Innate and following his blog if you want a better intuition for what is "obvious" to people in the field that outsiders tend to get totally wrong.

Beyond screening embryos out for Mendelian disorders, the next best likely scenario is to screen out really rare mutations that tend to higher impact/causal risk factor for developing a disorder.

The "sea of weak & maybe causal effects" problem is also generally true for several predictive models being developed in several areas of biomedicine and health. Prediction intervals for individual predictions can be obscenely large and rarely quantified. "Personalization" is not yet here.

There are many ways for disinformation. Like lumping a large set of other related data under some doctrine (e.g. equality of outcome) and dilute the effects aversed to their narratives. The selection effectiveness might be greater in higher IQ ranges as there will be more scope of IQ variation. By having massive IQ challenge mediocre parents it is alwas possible to declare that there are "Limit Utility".

The SAT score is calibrated from the national cognitive percentile. With the percentile the related IQ values can be estimated,

SAT = +13.74*IQsat -331.16; #n=357; Rsq=0.995; p=0 *** (VVSIG)

For 2.5 IQ point improvement, it is equivalent to 34.35 SAT score points. So on meritocratic ground based on the 50 percentile admission SAT scores, only a handful of enhanced students form tier1 universities can hop on to Harvard.

IQdIf | SAT50 | Inst
30 | 1475 | Stanford University
25 | 1480 | Harvey Mudd College
20 | 1485 | Columbia University
5 | 1500 | Massachusetts Institute of Technology
0 | 1505 | Princeton University
0 | 1505 | Harvard University
0 | 1505 | Yale University

However, many embryo selections, next to disease avoidance, some might be related to economic advantages. The economic amounts involved are non-linear and are higer at higher IQ ranges. By assuming "equality of outcomes" and lumping in large numbers from that cesspool any advantages in the high IQ ranges will be diluted. Thus the paper's assertion is intentionally biased.

From the data of New York Fed Reserve

and the uni majors entry IQ data from EduTestServ, the body that conduct the GRE tests.

It has been previously shown that for IQentry ≤ 115, on average the higher the IQentry, the higher the "under-employment rate", i.e. the rate where graduates are working in jobs that do not require any uni degrees, i.e. the meritocray proecesses are totally screwed up, if not on meritocary, competitent students have to compete with the sub-par students (which should be be there but thanks to "equality of outcomes") on wokeness, back steppings, etc. The progression this year is that some actually "woken" up and going for those STEM places and that had weaken the IQentry effect at the higher ends.

IQentry≤115: UnderEmp=+1.6*IQentryLo-137; #n=28; Rsq=0.106; p=0.0908 . (NotSig)\n\
IQentry>115: UnderEmp=-1.27*IQentryHi+195; #n=38; Rsq=0.133; p=0.02454 * (Sig)

IQentry≤115: MedianEarlyK = +0.49*IQ -17.95; #n=23; Rsq=0.055; pval=0.281 (NotSig)
IQentry>115: MedianEarlyK = +2.02*IQ +207.19; #n=34; Rsq=0.412; pval=4.347e-05 (VVSig)

Then suddenly in the mid career levels sanity returns, the MedianMid Salaries are again directly and possitively influenced by IQentry, those rift raft had been filtered out.

IQentry≤115: MedianMidK = +2.229*IQentryLo -185.967 ; #n=28; Rsq=0.235; SEest=1.352e+04; p=0.008934 ** (VSig)
IQentry>115: MedianMidK = +2.713*IQentryHi -258.834 ; #n=38; Rsq=0.4338; SEest=1.213e+04; p=6.937e-06 *** (VVSig)

Thus even with the same IQ gain of 2.5, on average it has no effect on MedianEarly salary for IQentry≤115 but otherwise the economic gains are considerable. Thus the paper's assertion is not valid in the economic sense.

The marginal IQ gains in IQentry>115 range can be obvious in some real situations. With IQgain of 2.5 point the students might be able to enter more gainful employments. With IQ gain of 2 (in real world IQ scores are without decimal point),for example, a Biology major (IQentry=121) with potential MedianEarly salary of only $35K might be able to hop to the Industrial Engeineering major (IQentry=123) with MedianEarly salary of $64K per annum, a gain of $29K per annum.

And in mid career Biologist MedianMid salary $63.2K, even less than the MedianEarly salary for Industrial Engineer, and whose MedianMid salaryr $87K. Does that follows from the paper's assertion of "Limited Utility"?

NPV of the gain for the Biologist and Industrial Engineer assuming current inflation rate of 1.8%, back to time of conception, assuming start work at 25 yo, entering MedianMid career salaries at 35
Ind ENgr return from 25+ yo NPV return=$1,449,491, NPV gain=$1,429,491.
Biologist return from 25+ yo NPV return=$985,826, NPV gain=$985,826.
The difference came from the on average potential IQ qain of 2.5 points and embryo selection cost $20K.

The above is not one off example. There are many other example of over $10K gain in Salaries for the "mere" 2 IQ gains,

$Diff | IQ1 | Major1 | MedianEarly1 | IQ2 | Major2 | MedianEarly2
32000 | 123 | Industrial Engineering | 64000 | 121 | Theology and Religion | 320
29000 | 123 | Industrial Engineering | 64000 | 121 | Biology | 35000
24000 | 126 | Electrical Engineering | 65000 | 124 | Chemistry | 41000
24000 | 123 | Industrial Engineering | 64000 | 121 | Earth Sciences | 40000
24000 | 123 | Industrial Engineering | 64000 | 121 | Biochemistry | 40000
23000 | 126 | Aerospace Engineering | 64000 | 124 | Chemistry | 41000
22000 | 123 | Industrial Engineering | 64000 | 121 | Geography | 42000
22000 | 126 | Mechanical Engineering | 63000 | 124 | Chemistry | 41000
21000 | 111 | Business Analytics | 57000 | 110 | General Education | 36000
20000 | 111 | Business Analytics | 57000 | 109 | Miscellaneous Education | 37000
20000 | 111 | Business Analytics | 57000 | 109 | Special Education | 37000
19000 | 126 | General Engineering | 60000 | 124 | Chemistry | 41000
18000 | 128 | Chemical Engineering | 68000 | 126 | Engineering Technologies | 50000
17000 | 111 | Business Analytics | 57000 | 109 | Public Policy and Law | 40000
16000 | 114 | Nursing | 50000 | 113 | Psychology | 34000
15000 | 126 | Electrical Engineering | 65000 | 124 | Information Systems and Management | 50000
15000 | 110 | Accounting | 50000 | 108 | Elementary Education | 35000
14000 | 126 | Aerospace Engineering | 64000 | 124 | Information Systems and Management | 50000
14000 | 130 | Mathematics | 50000 | 129 | Philosophy | 36000
13000 | 126 | Electrical Engineering | 65000 | 125 | Finance | 52000
13000 | 126 | Mechanical Engineering | 63000 | 124 | Information Systems and Management | 50000
13000 | 110 | Accounting | 50000 | 109 | Miscellaneous Education | 37000
13000 | 130 | Mathematics | 50000 | 128 | Miscellaneous Technologies | 37000
12000 | 126 | Aerospace Engineering | 64000 | 125 | Finance | 52000
12000 | 120 | International Affairs | 45000 | 119 | Anthropology | 33000
11000 | 126 | Mechanical Engineering | 63000 | 125 | Finance | 52000
11000 | 125 | Finance | 52000 | 124 | Chemistry | 41000

Links been weirdly blocked twice (or more) by the filters, but it is worth checking out gwern's page on embryo selection (google->gwern embryo selection, first link). Roughly comes to the same conclusion:

a meta-analysis of GCTA results indicates that SNPs can explain >33% of variance in current intelligence scores, and >44% with better-quality phenotype testing this sets an upper bound on the effectiveness of selection: a gain of 9 IQ points

when selecting the top embryo out of 10 the best 2016 polygenic score could achieve a gain of ~3 IQ points when selecting out of 10</i).

With discussion of almost everything you could think of relating to this topic.

I'm a little skeptical of the methods, and think the results are larger, but even if there are relatively small gains for extremely polygenic traits, they can be reliably ratcheted up by a little each generation. In other words, if every couple were doing embryonic screening, after 3 generations, it is plausible that you will have screened > 250 embryos, from a constantly improving stock. And that's just assuming each couple only has two children. It should be thought of as being similar to compound interest.

So there's no such thing as a genetic mutation.

To me, this is huge. Hopefully the whining of statist dorks like Ewan Birney don't delay this awesomeness for too long.

Comments for this post are closed