Polygenic screening is a method for modifying the traits of future children via embryo selection. If that sounds like gobbledygook, then think of it a bit like choosing stats for your baby.

That may sound amazing. It may sound like science fiction. It may even sound horribly dystopian. But whatever your feelings, it is in fact possible. And these benefits are available right now for a price that, while expensive, is within reach for most middle-class families.

On a more serious note, there is limited selection power available with today's technologies, so you will not be able to have a baby Einstein unless you are already a Nobel laureate. But polygenic screening will allow you to decrease your child's risk of common diseases by 10-60%, reduce their risk of mental disorders, and increase their IQ by somewhere between 3 and 8 points. If you are willing to wait a few years, you may be able to increase IQ by up to 13 points. Including the cost of IVF and testing, these benefits are available for between $30k-100k depending on the mother's age, how strong of a benefit you want and what kinds of traits you want to select for.

There has been quite a bit of discussion of this topic on LessWrong and adjacent communities but very little concrete advice for would-be parents who are curious whether the benefits are worth the price, particularly for those who have no other reason to do IVF. The purpose of this post is to fill that gap by addressing costs, potential medical complications, choice of clinic, which labs are best, and how age and infertility diagnosis affect the expected benefits.

This is a long post and I expect most people will not want to read the whole thing. If this is you, please use the section selector in the sidebar to navigate to the section you are most interested in. You may want to simply skip to the section titled "The Benefits of Polygenic Embryo Screening".

Background on IVF

A diagram showing the steps in an IVF cycle. Source

Wait, what even is polygenic embryo selection?

Embryo selection is all about picking an embryo to (hopefully) turn into a baby. This occurs during the process of In-Vitro Fertilization, or IVF. In the typical IVF cycle, a couple goes into a fertility clinic because they want to have a baby. Usually this is because they’ve been having trouble conceiving naturally, but couples also seek out IVF when they want to do genetic testing, select the sex of their child, or to preserve fertility for later pregnancy.

The doctor conducts a bunch of medical tests, and if they all check out, the woman begins a hormone regimen that will stimulate an abnormally large number of her eggs to mature all at once.

At the end of the regiment, the doctor extracts a bunch of mature eggs from the woman's ovaries, which are then fertilized using the father's sperm and grown in a lab dish for 4-7 days. When the embryo has finished growing, there are often four or more that can be implanted in the mother. Most couples do not want four children, so a choice must be made about which embryo to pick.

In ye olden days, doctors would often just transfer all the embryos at once in the hope that at least one of them would result in a baby. Sometimes this would work well; one of the embryos would happen to stick and the parents would be very happy. Other times it would work a little too well and more than one of the embryos would implant. This is why twin births are so much more common during IVF than during normal pregnancy.

Multiple birth rate in the UK over time. The trend is broadly similar in the USA. Source

Transferring multiple embryos at a time is less common nowadays because IVF clinics have figured out how to reduce the odds of failed pregnancy using genetic testing. With a higher chance of live birth from a single embryo transfer, the risk of a failed embryo transfer is outweighed by the risks of a twin pregnancy. The outcomes for twin births are on average worse than for single pregnancies. Twins are more likely to be born preterm, develop health problems, and put excess stress on the mother's body.

This brings us back to embryo selection; the doctor or embryologist has to make a choice about which embryo to transfer first. All clinics have to make this choice, so all practice embryo selection of some kind. But the criteria for selecting the embryo have, until recently, been pretty dumb.

The standard practice is for an embryologist to look at all the embryos under a microscope and pick the one that looks the prettiest. I am not kidding. The embryologist will rank the embryos from best to worst based on their "morphology", which accounts for factors like their rotational symmetry and whether or not they have a dark and rough colored appearance.

This embryo is pretty and will be transferred first
This embryo is ugly and will only be transferred as a last resort

To be fair to the embryologists, this method is better than just randomly picking an embryo; embryos with particularly bad morphology gradings do actually have a lower chance of resulting in a live birth. And for a long time, there was simply no other option But times have changed and we can now select embryos by DNA rather than simply the appearance of their cells under a microscope.

But how do they even get an embryo's DNA?

The Blastocyst Biopsy - RGI
This is what an embryo biopsy looks like. You can see a video of the process here

All the best techniques for genotyping rely on destructive sequencing, meaning the cells whose DNA is read must be destroyed. Embryos don't have very many cells. So how do we get information about what's in its genome without destroying it?

It turns out that after roughly five days of development, embryos possess a very interesting property; one may remove up to about 10 cells with little to no measurable impact on the embryo's ability to develop into a healthy child. The embryo can regenerate up to about 10% of its mass! That’s the equivalent of losing and then regrowing both your arms as an adult. This is very fortunate for us because these cells contain a treasure trove of information.

The most common thing IVF clinics look for is aneuploidy, which is a medical term meaning "this embryo has an abnormal number of chromosomes". The term for this type of testing is “PGT-A”, and it’s performed in roughly half of all IVF cycles in the US today.

Human embryos with the wrong number of chromosomes are surprisingly common, both among IVF patients and natural pregnancy. But this wasn’t very well understood before the first use of pre-implantation genetic testing in the late 1980s.

IVF Doctors started wondering why so many transfers were failing to result in pregnancy, or resulting in pregnancy followed by very early miscarriage. They discovered that roughly a third(!) of all pregnancies, both natural and via IVF have chromosomal abnormalities. Most of the time these go undetected because their immediate effect is to result in the arrest of the embryo’s growth, or to cause a very early miscarriage (often before the woman even knows she is pregnant).

IVF clinics began testing embryos for aneuploidy in the early 1980s. But in the late 2000s, something happened that completely changed the landscape of genetic testing.

Why Polygenic Screening wasn't possible before 2015

The cost of sequencing a fixed amount of DNA has declined dramatically since Fred Sanger and his team pioneered the first methods in 1977. There is a kind of "Moore's law of sequencing" in which the cost of sequencing a fixed amount of DNA has declined exponentially over time.

However, something incredibly dramatic happened to DNA sequencing in about 2007. Take a look at this graph:

I don't think I've ever seen a graph that looks like this anywhere else. Between 2007 and 2010, the cost of sequencing a megabase of DNA dropped by a factor of a million! That unbelievable, super exponential drop was made possible by a technology called "Next Generation Sequencing".

By the mid-2010s, you could genotype all the parts of a person’s DNA most likely to differ from other people’s for under $100. At that price point, it became possible to gather genomes from hundreds of thousands of people and assemble them into giant databases that researchers could access. 

This was incredibly important, because you NEED hundreds of thousands of samples to make good genetic predictors. It turns out that most of the traits we care about like heart disease risk or intelligence or attractiveness are determined not by a handful of high-impact genes, but by the cumulative effect of thousands of genes, each of which has only a tiny impact.

Take educational attainment. Educational attainment is not the most heritable trait, but because research on it is more politically acceptable in a university environment than the direct study of intelligence, we know quite a bit about its genetic roots. The latest large-scale study of it included data from 2.7 million participants. Among all genes identified, the one with the single largest effect size only increased the amount of time you spent in school by at most 2.8 weeks (see section 3.4). That's it! The average gene has a tiny, tiny impact on how long you spend in school. The predictor used 2,925 genes to explain just 15% of the variance in how many years of school a person completed.

So you actually NEED these gigantic databases to explain more than a tiny fraction of the variance in complex traits. This is why polygenic embryo selection was impossible before about 2016; there just wasn't enough data to figure out which genes did what.

How do they know which genes do what?

Once you have a giant biobank and information about people's traits and diseases, you still need to figure out which genes do what. I mentioned an educational attainment predictor in the section above, but I didn't explain how they created it. So how did they do it?

The answer is actually not too complicated: a researcher will use one of these gigantic biobanks plus a machine learning model to identify which genetic variants are associated with an increase or decrease in a given trait. 

The dumbest possible way to do this is with a Genome Wide Association Study, or GWAS. It works something like this:

Let's say there's a gene with two different variants commonly present in the biobank population. 96% of participants have an "A" at some particular location in the gene, but 4% have a "T" instead. We want to know whether having a "T" makes you taller.

A GWAS just measures the average height of people with an A and the height of people with a T to see if they're different. Then it uses a statistical significance test to see if the result could have plausibly been the result of random chance. If not, the researchers reason that the gene is having an effect on height. If it passes this test, it is added to the "list of important genes for height".

For such a dumb method, this works remarkably well. Height predictors created using GWAS results correlate with actual height at about 0.55

The smarter way to do this is to use some kind of machine learning method like LASSO. This will give you a better predictor for the same amount of data. But for some reason I still don't really understand, academia almost exclusively uses GWAS.

A graph showing LASSO predicted vs observed height in a validation cohort.

Correlation or causation?

"OK", you might say. "That's well and good, but how do we know that these genetic differences are actually CAUSING someone to be taller or smarter rather than just spuriously correlated with height?"

The main reason this is possible is because nature has already conducted a randomized control trial on our behalf. Every time your body produces a sperm or egg cell, your DNA is more or less randomly mixed up and half of it is given to the reproductive cell. This means that, conditional on parental genomes, sibling genomes are randomized!

A diagram showing how genetic randomization (meiosis) works for a single chromosome.

In turn, this means that if a gene can predict differences between siblings, you can be quite confident that it is in fact CAUSING the difference. This is actually quite a remarkable fact, and one that underpins the entire reason for believing embryo selection should work.

There is one asterisk here; though a sibling GWAS can tell you where the causal variant is, it usually can only narrow down the list of candidates to perhaps 10 distinct variants within a region of very roughly 100,000 base pairs. This is sufficient for embryo selection because that set of 10 base pairs will almost always be inherited together. But if sometime down the line we want to do embryo editing, it will require us to either pinpoint the causal variant precisely or to edit all 10 variants that have a decent chance of causing the observed change.

Another crucial insight from these studies is that nearly all of the genetic differences between humans can be explained by additive effects; there are very few gene-gene interactions going on; If gene A makes you taller, it doesn't depend on gene B being present to work its magic. It's a strong, independent gene that don't need no help.

This fact is extremely important because it makes both evolution and embryo selection possible. There is a common misconception that genes are tied together in a hopelessly complex web and that if we mess with one part of it the whole thing will come crashing down. While that may be true for genes that are universally present in the human population, it is very rarely true for genes that commonly vary between people.

You have the predictors. Now what?

Once you have created genetic predictors using GWAS or LASSO or some other method, you can then feed in the embryo's DNA to the trained model and get a prediction of each embryo's expected trait value. You do this for every predictor you have (or at least those you care about), and then pick an embryo based on the results.

But there's one last question to answer: which traits should you care about? If one embryo has a 20% chance of getting breast cancer and a 10% chance of getting heart disease, is that better or worse than an embryo with a 10% risk of breast cancer and a 20% risk of heart disease? Or how about one that has a high risk of both but is also predicted to have an IQ 5 points above average?

There is no universally agreed-upon method for making the choice about which embryo to implant. My personal hope is that someone (maybe even me!) makes a tool to assess what parents find important and then ranks embryos according to those criteria.

The Benefits of Polygenic Embryo Screening

CategoryTraitImprovement RangePublicly Available?
Non-diseaseIntelligence+1.6-7.5 IQ pointsNo*
Non-diseaseHeight1-6 cmNo*
DiseaseAlzheimer's15-50% reductionYes
DiseaseAtrial Fibrillation10-50%Yes
DiseaseBreast Cancer3-50%Yes
DiseaseBasal Cell Carcinoma3-45%Yes
DiseaseCoronary Artery Disease20-60%Yes
DiseaseHeart Attack25-70%Yes
DiseaseHigh Cholesterol12-50%Yes
DiseaseInflammatory Bowel Disorder5-65%Yes
DiseaseIschemic Stroke5-20%Yes
DiseaseProstate Cancer2-60%Yes
DiseaseType 1 Diabetes10-55%Yes
DiseaseType 2 Diabetes20-60%Yes
DiseaseTesticular Cancer0-55%Yes
Mental DisorderMajor Depressive Disorder5-20%Yes
Mental DisorderSchizophrenia5-75%Yes

*See the section below for how to get access to these predictors

Ok, enough with the theory. How big of a benefit can you actually get from going through IVF and screening your embryos?

I'll start with the one everyone always asks about: intelligence. How much can you boost your child's IQ with embryo selection?

How much can embryo selection increase my child's IQ?

First, there is no company that publicly offers embryo selection for intelligence. I have spoken with a stealth mode startup that offers selection for disease and non-disease traits, including intelligence. If you're interested in selecting for non-disease triats, you can get in touch with them via Jonathan Anomaly, who knows some of the people working at the company.

Their current predictor correlates with measured IQ at about 0.4, which means they've likely compiled data from multiple sources to create it.

So how big would the gain be? Using some code from Gwern's monster post on embryo selection for intelligence and some results from their calculator, I created the following graph:

Expected IQ gain as a function of achievable births for parents of European ancestry (note this is assuming you select only for IQ)

It's plausible that you would get up to maybe 40 euploid embryos if the mother is young and you do multiple rounds of egg retrieval. In that case, you could probably get a gain closer to 8 points. If the mother is older it will be less. There's also a reduction in benefit if one of the parents is of non-european ancestry: likely around 8% for Ashkenazis, and 20% for east asians, and probably a similar or lesser reduction for Indians. I am uncertain of the reduction in gain for those of African ancestry, but it would likely be larger (perhaps 30-40%?).

This is an unfortunate side-effect of the fact that there aren't enough non-Europeans in the large biobanks on which these predictors are trained.

There is significant room for this benefit to improve in the near future. The million veterans project in the US has whole genome sequences and ASVAB test scores for (you guessed it), a million soldiers. If researchers were simply allowed to use this existing data to create an intelligence predictor, the gain from embryo selectioin would increase to 8.5-13 IQ points and the racial disparities in predictor quality would mostly disappear. The marginal cost of this would be virtually zero.

This future increase in the efficacy of embryo selection has an obvious implication: if you freeze eggs or embryos now, you'll have more embryos to pick from (since egg production declines with maternal age), and if you wait a few years to implant them, the expected gain from selection will be higher.

Disease Reduction

Unlike intelligence, there actually are several companies that offer polygenic embryo screening for disease risk. For this reason, I can tell you quite a bit about exactly how much you can reduce disease via embryo selection.

There are two main "categories" of disease risk to think about: monogenic disease risk (which includes diseases like Tay Sachs, Cystic Fibrosis etc), and polygenic disease risk, which includes heart disease, alzheimers, schizophrenia, diabetes and most others.

I'm going to focus on polygenic screening, since everyone has a non-zero risk of them, and embryo selection to reduce polygenic disease risk is already available in clinics.

The first company to offer this was Genomic Prediction. Orchid Health also finally offers polygenic embryo screening, though I believe they charge more per embryo. One other lab in China appears to have recently deployed a similar test, though it looks as though they screened embryos for type 2 diabetes risk exclusively, which is not a very sensible strategy in my opinion. And lastly, there is a stealth mode startup doing polygenic embryo screening. They offer screening for both disease risks and non-disease traits like intelligence. If you're interested in contacting them you can reach them through Jonathan Anomaly.

 Because I have little information about the predictor quality of Orchid Health or the stealth mode startup, I will focus on papers published by Genomic Prediction.

They use a pretty straightforward and simple method for determining an embryo's relative ranking: each disease is weighted according to its impact on disability-adjusted lifespan. According to one of their recent papers, selecting embryos in this manner results in a fairly impressive reduction in disease risk across multiple conditions. Here's a graph from one of their latest papers showing the expected reduction in the relative risk of various diseases from selecting the best of five embryos.

AD = Alzheimers, AFib = Atrial Fibrillation, ASA = Asthma, BC = Breast Cancer, BCC = Basal Cell Carcinoma, CAD = Coronary Artery Disease, HA = Heart Attack, HCL = High Cholesterol, HTN = Hypertension, IBD = Inflammatory Bowel Disorder, IS = Ischemic Stroke, MDD = Major Depressive Disorder, MM = Malignant Melanoma, Obes = Obesity, PC = Prostate Cancer, SCZ = Schizophrenia, T1D = Type 1 Diabetes, T2D = Type 2 Diabetes, TC = Testicular Cancer

The figure above shows disease reductions for selection among 5 "pseudosiblings" of European descent (apparently there are not enough real siblings to train predictors on). My guess is the benefits will be reduced compared to those shown, perhaps by around 20%.

Note also that this doesn’t take losses from implantation or miscarriage into account. So you’ll need more like 7 or 8 euploid embryos to achieve gains 20% lower than this. But still, the benefits are reasonably strong.

The most amazing thing to me about the above graph is that selection seems to reduce EVERY disease in the index. One of the major concerns I hear raised about embryo selection is that there might be some "hidden downside" to selection. This graph seems to suggest that, at least so far as diseases go, that's not much of a concern.

Another way of looking at the benefits of disease reduction is to look at how much the quality-adjusted lifespan of an average child born via polygenic screening would increase compared to one born without its benefits. Here's their analysis of this framing:

Just replace “Group size” with “number of achievable births” in your head

Keep in mind, this is the pseudosibling benefit, so take what you see on the graph above and multiply it by 0.8 to get a more realistic number. Also, the benefit is not quite as large for non-European groups. It looks like the gain is reduced by about 20-30% for South Asians, 30% for Africans and 35% for East Asians.

Of course, this all kind of ignores the elephant in the room: many of these diseases have an average age of onset of 50-75. In fifty years the world is likely to look incredibly different. If humans are still around, it seems likely that we'll have cures or at least very effective treatments for many of these conditions.

The exceptions are mental disorders like clinical depression and ADHD, which generally have an average age of onset before age 25, and obesity, which is now showing up more and more in childhood.

If Orchid Health is more expensive, why would anyone use their tests?

Orchid Health offers whole genome embryo sequencing. That means instead of just looking at ~800,000 spots in the genome, they look at about 3 billion.

This doesn't make as big of a difference as you might think; most of the information we care about is already present in one of those 800,000 spots that companies like Genomic Prediction examine with their embryo testing. But whole genome sequencing has one specific advantage that companies like Genomic Prediction can't replicate: they can detect de novo mutations.

De novo mutations are random genetic mutations that aren't present in either of the parents. When you think of "random mutation and natural selection", the "random mutation" part IS de novo variants.

We hear a lot about mutations conferring evolutionary advantages, both in science class when learning about evolution and in science fiction stories like X-men, but unfortunately most de novo mutations are just bad for you. Some of them are REALLY bad, as in your child will die quickly after birth or live a permanently stunted life.

Orchid's testing finds one of these monogenic issues in about 3.7% of embryos. I'll update this post later with more concrete info about the size of the benefit from having a child without those problems, but at a high level I think it's likely worth the price if you can afford it.

Other Non-Disease Traits

Personality Traits

As with intelligence, there are no companies publicly offering screening for personality traits to the best of my knowledge. The stealth mode company has told me they may offer personality prediction at some point, though I am uncertain of their quality. 

Predictors for personality traits such as conscientiousness and neuroticism are poor. This seems to be partly related to the lack of good data on such traits. A paper published in March of 2022 was able to explain 2% of the variance in conscientiousness and much less for others. There is significant room for improvement here, as most estimates peg the heritability of the big five personality traits at roughly 40-60%.

There are existing third-party services like Genomelink and SelfDecode which provide personality predictors. Unfortunately, they don't have any published data on the quality of the predictors they use, and the anecdotes I've heard suggest that, at least for intelligence predictors, they are very poor. So if you're looking for something that isn't terrible, your best bet is to contact Jonathan Anomaly and hope the stealth mode startup offers something better.

Facial Features/Attractiveness

It would be nice (or possibly dystopian, depending on your views) if you could add physical attractiveness to your embryo selection criteria. More attractive people have higher lifetime income, better dating prospects, and seem to benefit generally from the halo effect.

To the best of my knowledge, no one offers this as a service right now, so unless the super secret startup offers it, you'd probably have to pay a researcher tens of thousands of dollars to develop a predictor for you.

The predictor might be halfway decent, though I don't expect it to be as good as the ones we have for intelligence or most diseases. Here's a meta-analysis of facial feature genetics from 2020 that found 203 genome-wide significant signals. If someone made a predictor using that paper it might be able to predict facial attractiveness reasonably well. However, there is clearly much room for improvement.

A neat picture from the meta-analysis linked above showing how some of the top SNPs affect facial structure.

Is attractiveness a purely position good? That is to say, if everyone was made attractice, would society be better off? My guess is probably a little, but I haven’t done enough research into this topic to be sure. 

If you want to see what selection on attractiveness looks like in the extreme, take a look at male birds of paradise. Female birds have been selecting mates based on their plumage and dancing ability for thousands of generations, resulting in some very beautiful, but very strange looking birds. 

What’s the human equivalent of these birds?

It is worth asking ourselves if dozens of generations of selection for attractiveness might result in similarly strange and pointless features whose sole purpose is to elicit arousal in members of the opposite sex. 

And in the shorter run, selection for physical attractiveness probably trades off at least somewhat against selection for other features that have a greater benefit for society as a whole. So for now there is a non-zero cost to making everyone more attractive.

Still, from a purely selfish perspective, some level of selection for attractiveness makes sense. As with intelligence, no company currently offers the ability to select for physical attractiveness, unless you include selection for height, which already works extremely well. Your best bet is contacting Jonathan Anomaly to see if the groups he knows offer something. Or, barring that, you could hire a PHD student for 50k and pay them to develop a predictor for you.

Concrete Advice for Would-Be Parents

If you've read the sections above and you decided you want to do polygenic embryo screening, your primary objective should be to get maximum gain for minimum cost. This section is about how to achieve that goal.

There are two primary inputs that determine the effect size of polygenic screening:

  1. The power of the polygenic predictors used to select an embryo for implantation.
  2. The number of "achievable births", meaning the number of children you would have if you implanted all of your embryos one by one.

Number 1 is largely out of your control unless your name is Gwern or you are a research scientist with access to a large biobank.

Number 2 IS within your control, at least to some degree. Here is a list of factors you might be able to change that will influence achievable births:

  1. How many IVF cycles you go through
  2. The IVF clinic you choose
  3. The PGT lab used for aneuploidy testing
  4. The age of the mother or egg donor at the time eggs are extracted (younger is always better)
  5. The stimulation protocol used
  6. The age of the father or sperm donor (younger is better)
  7. Whether you choose to freeze eggs or embryos
  8. Routine prenatal care that any good obstetrician will be able to tell you (though see Emily Oster's excellent book if you want more details)

Why do these factors matter? Because at every step of the IVF process from initial consultation to birth, fewer eggs/embryos/pregnancies come out than go in. There's a loss associated with each step, and the factors listed above have a large influence on the size of the loss. Here are some suggestions on how to reduce losses and costs:

Suggestion #1: Reduce uncertainty about how many euploid embryos you will produce

The single most important input into the “gain” equation is the number of mature eggs harvested per retrieval. However, this quantity has a very wide distribution. In conversations an acquaintance of mine had with egg donor clinics, they mentioned that some donors produce as many as 100 eggs per retrieval. A woman in her mid-40s with infertility issues might produce 3.

There are several heuristics you can use to reduce the uncertainty about how many eggs you are likely to harvest during an egg retrieval. Knowing these beforehand can help reduce uncertainty about the exact size of the benefits of polygenic screening.

The easiest heuristic to use is the woman’s age. As a general rule, I wouldn’t do IVF for the purpose of polygenic embryo selection unless the mother is under 38. Beyond age 38, the losses in the IVF funnel are just too steep to justify the pain and expense, unless you plan to use donor eggs.

Another lower-cost but not free thing you can do is assess your likely egg production by undergoing the very first part of IVF; a consultation and ovarian ultrasound. This ultrasound is performed right at the very start of the IVF process and usually costs less than $1000. Ovarian reserve and antral follicle count are strongly correlated with the expected number of mature eggs you or your female partner will produce after hormonal treatments. If you’re willing to do embryo selection in theory if the gain is large enough, this can be a relatively inexpensive way to reduce the uncertainty about the benefits of the procedure.

Antral Follicle Counting with Ultrasound Video
An ultrasound to assess antral follicle count

What is really needed here is a tool that allows one to see the expected gain across a variety of traits given an antral follicle count. The data necessary to do this research is present in the CDC’s NASS database, but I don’t yet have access. I hope to do this for a future research project. Until then, here’s a table showing roughly how your odds vary as a function of antral follicle count.

Suggestion #2: Use my table to pick a clinic

I have spent an embarrassing amount of time on a research project to rank every IVF clinic in the US from best to worst. I compiled this list using data from the CDC’s NASS database, which has information about clinics going back to the mid-90s. I believe I am the first and only person in the world to do this. If you’re curious about a clinic that isn’t on the last, feel free to reach out to me and I can give you the numbers.

A somewhat boring explanation of the research I did (skip this if you just want results)

Why Does The Best Sleep Come In A Boring Lecture? » Science ABC
Warning: you may look like this student after reading about my research methodology

Clinics are ranked according to their cumulative live birth rate per intended egg retrieval among patients using their own eggs (not donor eggs). In simple English, that means we’re looking at what percentage of women who started hormone treatments actually delivered at least one child.

In an ideal world I’d give you clinic rankings based on the number of expected births per retrieval. But without access to intermediate outcome data from NASS, this is impossible. I plan to apply for access eventually, but in the meantime I think live birth rate is probably quite a good proxy, and clinics with very high live birth rates are likely to be able to produce more achievable births than those with lower live birth rates.

I’ve taken care to control for various factors that could confound the analysis. Some clinics attract a larger proportion of younger patients, who have better prospects than older cohorts. Some clinics attract a large proportion of patients solely interested in freezing eggs or embryos for some later treatment. Some clinics have a very small number of cycles each year and can score very well or very poorly depending on the luck of the draw. I’ve controlled for all of these confounders in my analysis.

The one big thing I didn’t control for was the percentage of patients presenting with a given infertility diagnosis. I attempted to to this in an earlier version of the project but had weeks and weeks of nightmares trying to find some defensible way to deal with the large amount of censored data present in spreadsheets. Supposedly the CDC censors these values to protect patient privacy. This justification is obviously nonsense; they still have uncensored values from before 2018 on their website. I emailed to ask if I could apply for some kind of special access as a researcher and was denied.

I tried to work around these missing vaues but eventually I simply gave up. I had to make too many unjustified assumptions to compute clinic rankings, and ranking was highly variable depending on which assumptions I made.

So this final analysis controls only for maternal age, use of patient’s eggs (vs donor eggs), percentage of retrievals conducted with the intention to freeze embryos or eggs (obviously those people aren’t going to have a baby), and some bayesian averaging sprinkled in on top to differentially bring clinics with low retrievals/year more towards the mean of all clinics.

I plan to actually publish my results in proper academic setting at some point, so this post contains only the headline numbers. 

Without further ado, here are the top 25 IVF clinics in the US as of 2020.

The best IVF clinics in the USA

Clinic NameAdjusted Live Birth  RateClinic StatePhone Number
Carolinas Fertility Institute0.516123North Carolina(336) 448-9100
The Georgia Center for Reproductive Medicine (no website)0.490339Georgia(912) 352-8588
Reproductive Gynecology & Infertility-Westerville (Columbus location)0.482645Ohio(614) 895-3333
Reproductive Medicine Associates of New Jersey0.458511New Jersey(973) 971-4600
Center for Reproductive Medicine, Advanced Reproductive Technologies0.457538Minnesota(612) 863-5390
Missouri Fertility0.456779Missouri(573) 443-4511
Spring Fertility (San Francisco location)0.446066California(415) 964-5618
CCRM Boston (main center in Chestnut Hill)0.439929Massachusetts(617) 449-9750
SpringCreek Fertility (Dayton location)0.437458Ohio(937) 458-5084
Duke Fertility Center, Duke University Medical Center0.419579North Carolina(919) 572-4673
New Direction Fertility Centers0.413924Arizona(480) 351-8222
Shady Grove Fertility Colorado0.411165Colorado(720) 704-8221
Center for Advanced Reproductive Medicine0.409579Kansas(913) 588-2229
Fertility Center of the Carolinas0.406704South Carolina(864) 455-1600
Fertility Center of San Antonio0.403583Texas(210) 692-0577
Baystate Reproductive Medicine0.403039Massachusetts(413) 794-1950
University of Iowa Hospitals and Clinics, Center for Advanced Reproductive Care0.401587Iowa(319) 356-8483
Carilion Clinic Reproductive Medicine and Fertility0.400715Virginia(540) 985-8078
Advanced Fertility Center of Chicago0.399771Illinois(847) 662-1818
Shady Grove Fertility-Richmond0.394682Virginia(804) 379-9000
Fertility Center of Southern California0.391824California(949) 955-0072
Northern California Fertility Medical Center0.390555California(916) 773-2229
The Nevada Center for Reproductive Medicine0.390344Nevada(775) 828-1200
Center for Advanced Reproductive Services (Farmington Location)0.388567Connecticut(844) 467-3483

For reference, the average adjusted live birth rate for all clinics nationwide was 0.278.

How predictive are an IVF clinic’s past success rates of their future success rates? Here’s a graph showing how well a clinic’s 2017 live birth rates correlated with their 2020 live birth rates after adjusting for the confounders I mentioned above:

The overall takeaway here is fairly clear: by selecting one of the top 25 or so clinics in the most recent year, you can increase your achievable births by perhaps 20-30%.

I have been working on a more advanced version of the model used to produce the results above which makes more efficient use of crappy, censored data and shows results for 2021. I also plan on releasing a free app to the app store making all of this data much more usable. This post will be updated when it goes live.

Suggestion #3: Use a good PGT lab

TL;DR Make sure you use Orchid Health or Genomic Prediction for aneuploidy testing. They very likely have lower false positive and false negative rates for embryo aneuploidy when compared with other PGT labs.

It is a little known fact that there is a significant difference in aneuploidy false positive and false negative rates between PGT labs. To the best of my knowledge, there have been no randomized control trials comparing PGT labs. The best we have are independently conducted retrospective cohort analyses. 

Unfortunately most of these analyses do not disclose which labs are being analyzed, making them completely useless for paients. However, I happen to know which clinics are which for one particular study submitted to ASRM in 2021. In this study, clinic A is Igenomix, clinic B is Genomic Prediction, and Clinic C is Cooper Genomics. 

There’s a lot of sort of random statistics thrown out in this presentation, but I’d like to focus on those most relevant to achievable births: aneuploidy rates, pregnancy rates and miscarriage rates (which they break up into early and late miscarriages in the study)

If you watch the video, you’ll see that “Lab B (AKA Genomic Prediction)” is either as good or significantly better than the other two labs in the study across virtually every metric. If you add up the impact of these metrics, here’s what the expected number of achievable births look like for each clinic from the study:

Note that Genomic Prediction is labelled as “LifeView” on this graph. LifeView is the name of their PGT platform

I don't know of any direct comparisons of Orchid Health with other PGT labs, but my guess is their false positive and false negative rates are at least low as those of Genomic Prediction. They use whole genome embryo sequencing, which means they retrieve a gigantic amount of data from each embryo. It's likely this amount of data reduces testing errors compared to low density NGS sequencing, but I am not certain about this.

Suggestion #4: If possible, freeze embryos instead of eggs

This one is short and sweet: if you have a choice, freeze embryos instead of eggs. At a good storage facility using vitrification, about 90% of eggs will survive cryopreservation. At that same storage facility, 99% of embryos will survive. So if you already know who you want kids with, freeze embryos instead of eggs for an easy 10% boost in achievable births.

Suggestion #5: Freeze eggs or embryos as soon as possible

Since expected gain increases the more achievable births you have, all tips for maximizing it revolve around increasing the number of euploid blastocysts you can produce during IVF. You can pick a good clinic, use a good PGT lab, freeze embryos instead of eggs, and follow good prenatal care guidelines. But at the end of the day, the single biggest input variable into the “gain” equation is the age of the mother. 

Here’s a graph from another research project I did showing the relationship between maternal age and number of eggs retrieved at three different clinics 

Note that the numbers here are somewhat pessimistic as they are among an infertile population. 

You can see there’s more or less a linear decline in expected egg count per retrieval as a function of maternal age. It’s much the same story for expected zygotes and blastocysts; a linear or even exponential decline as a function of maternal age.

If you decide to do polygenic embryo screening, the sooner you start the process the better.

A compendium of other advice

  • When choosing a clinic, there’s several things you need to ensure they are OK with before you agree to become a patient
    • You need to be able to send your embryo’s biopsies to a lab of your choosing, or they need to already be working with Genomic Prediction or Orchid Health.
    • If you want to screen for anything other than the diseases offered by the above companies, the clinic must to be willing to implant an arbitrary embryo of your choice.
  • Make sure that the clinic you choose either has reasonable embryo storage costs or will let you ship your frozen embryos to a facility that does. Some clinics charge up to $1500/year for embryo storage and will raise the price on you as time goes on. Cheaper clinics charge under $1000/year for storage (some as little as $500)
  • Read Emily Oster’s excellent book about the things you should and shouldn’t do before, during and after pregnancy. Seriously, Oster is excellent and enjoyable to read. Ex: gardening is dangerous for pregnant women due to soil microbes but <3 drinks per week seems to be completely fine. The one possible exception to this is advice about alcohol consumption, but the science there is very complicated and confounded by selection effects.
  • If you’re a man over 30, consider freezing your sperm. De-novo mutations disproportionately cause conditions like cognitive deficits, severe autism and other serious conditions, and most of these mutations come from the man's sperm.

The IVF Loss Funnel

Ok, if we put together all the data above, how many live births can you get per egg retrieval?

IVF loss funnel for an average IVF patient

Naively interpreted, the above graph would imply that an average IVF cycle would only yield a single achievable birth on average. This is true! The average IVF patient is a 36 year old woman with significant fertility issues, so it’s not particularly common for such individuals to have more than one birth per egg retrieval. In fact, over half of IVF cycles do not result in live birth. The average is dragged up somewhat by the fact that some women are able to have multiple children from a single egg retrieval.

But what if you choose a better IVF clinic and a better PGT lab than average? What if you and your partner have no known infertility issues and your female partner is younger than the average 36 year-old IVF patient (say 32 for this example).

In that case, we'd expect the graph to look something like this:

IVF funnel for a woman in her early 30s with no infertility issues

Looking better! With a younger mother, a top-tier clinic, and no history of infertility, roughly 5 achievable births per retrieval is possible.

How about in the best case scenario? Assume the following:

  • The mother is at peak fertility (early to mid 20s)
  • Egg retrieval is performed using conventional IVF hormone treatment
  • The father is young-ish (under 40)
  • Neither parent has any infertility issues
  • The parents use a top-tier clinic that is very good at culturing eggs into blastocysts
  • They use a top-tier PGT clinic with a very accurate aneuploidy test
  • The eggs are fertilized immediately and the resulting blastocysts were frozen

In that case, things start to look A LOT better.

IVF funnel for a woman in her early to mid 20s with no infertility issues

The numbers above are based on a combination of sources including SART data on miscarriage and transfer loss rates, podcast episodes, publicly available data from egg donor clinics, and my own knowledge of the IVF industry. I suspect that it may be somewhat conservative for couples without infertility since I have used the infertile transfer and miscarriage rates. But they should nonetheless give a fairly accurate view of the loss funnel.

One last topic I shoud address; how much will all of this cost?

How much does IVF and PGT cost?

IVF is expensive. To do polygenic embryo screening you’ll need to pay for a consultation, ultrasounds, transportation to and from the clinic, IVF services like lab techs, medication, pre-implantation genetic testing, and data analysis services to select for non-disease traits.

I’ve called a couple of dozen top-tier IVF clinics on the phone to ask about prices. I’ll give you a general cost estimate based on those calls:

ServicePrice RangeModal Price
Follicular Ultrasound$150-500$400
Egg retrieval (not including transfer)$6000-20,000$14,000
Embryo Transfer$3000-$6000$4000
PGT-P$1500-5000$1000 + $400/embryo
Selection for intelligence, height, etc$15,000-$50,000Best guess: $20,000
Total$9000-$35,000$26,500 + intelligence screening

This process is not cheap. If you want to do two egg retrievals and select for intelligence you’re looking at a minimum of $50k and possibly much higher depending on what you'd like to select for. I sincerely hope we can bring the price of these services down in the coming years.


Polygenic embryo selection can currently increase your child's quality-adjusted life expectancy by 1-4 years, decrease their risk of various chronic diseases by 10-60%, increase their IQ by 2-8 points,  increase height by up to 2.25 inches, and moderately improve other traits. The exact gain you can expect to get for each of these traits varies depending on the genetic correlation between the traits, the number of embryos you have to choose from, and the strength of the predictor used to select embryos, as well as simple luck. Subsequent children will see a somewhat smaller but still positive benefit, though for every child to benefit you will need at least 3x the number of euploid embryos as you want children.

To get these benefits, you will have to go through IVF and genetic testing of your embryos, which will cost $20k-$60k (and perhaps more depending on whether you want custom testing) and require the female partner to take 2-6 weeks off work.

You can increase the expected gain by choosing a good IVF clinic, choosing a good PGT Lab, freezing embryos instead of eggs, and beginning the process as soon as possible since younger mothers produce significantly more eggs than older mothers.

The IVF process is not particularly pleasant, and is expensive to boot, if you and your partner are willing to put up with the discomfort and expense it you can give your children advantages that are impossible to get any other way.

If you freeze embryos now, the expected gain will increase over time as the genetic predictors used to select embryos improve, and the panel of traits which you can select for will also increase.

There are technologies on the horizon that will allow for significantly greater gains across all heritable traits, making possible gains of 4 or more standard deviations across multiple traits simultaneously.

If AI doesn't destroy the world first, the next 30 years will likely see the greatest crop of geniuses and athletes the human species has ever produced. If we are wise and select for traits like kindness, altruism, and happiness in addition to health, attractiveness and intelligence, the children born with these benefits may be able to guide the human species through the incredible upheaval and instability we are likely to see over the next century.


New Comment
109 comments, sorted by Click to highlight new comments since:
Some comments are truncated due to high volume. (⌘F to expand all)Change truncation settings

Thank you, this is a great post. A few questions:

  • You say "see below for how to get access to these predictors". Am I understanding right that the advice you're referring to is to contact Jonathan and see if he knows?
  • I heard a rumor that you can get IQ out of standard predictors like LifeView by looking at "risk of cognitive disability"; since cognitive disability is just IQ under a certain bar, this is covertly predicting IQ. Do you know anything about whether this is true?
  • I can't find any of these services listing cost clearly, but this older article https://www.genomeweb.com/sequencing/genomic-prediction-raises-45m#.ZFqXprDMJaR suggests a cost of $1,000 + 400*embryo for screening. Where did you get the $20,000 estimate?
Yes. My understanding is he knows some groups that have a working IQ predictor and are accepting customers. Genomic Prediction no longer offers an intellectual disability predictor. They got huge blowback when they first released that predictor and removed it from their traits as a result. I do not believe that you'd expect to get much of an IQ bump from selecting against disease risk either. My guess is less than 1 point from 10 achievable births. Sorry, I should really go back and edit the post to make this clearer. To the best of my knowledge Genomic Prediction (and possibly Orchid) are the only companies that can genotype your embryos with reasonably good quality and will give you the raw data. This is the part that (probably) costs about $1000 + 400 per embryo. You then have to take that raw data to a third-party service (probably one of the groups that Jonathan knows) and ask them to predict the IQ and/or other traits. I don't know anything about the groups, but I'd be shocked if they're doing this for free. So they will charge an additional amount, which is where my estimate of $20k came in. I don't actually know anything about their prices so that's a complete shot in the dark for what it costs. But given it's a low-volume service at this point, my guess is it's quite expensive.
5Scott Alexander

Does anyone here have qualms about the moral status of the embryos that are discarded in this process? I'm aware of the OP's views on the issue, and I recently addressed them elsewhere, but I'm curious about the average viewer of this page.

6Seth Herd
Emphatically no. They are primarily potential lives, but they need their mother's womb to achieve that potential, and they can't all have that. Those simple brains have no more inherent moral worth than an insect with a similar number of neurons. Only one life can come out of this process. Whether that one is chosen by chance or choice, the one resulting life has the same moral worth.
0Jakub Supeł
The statement that they are potential lives is incorrect. An embryo is already alive and, since it has continuity through time with an adult human being (obviously actual living human), it has human identity as well. Therefore, it is a living human being.   "Only one life can come out of this process" is also incorrect. This is like having 4 teenagers and choosing 3 of them to be shot, and then concluding that "only one adult can come out of this process, therefore the 3 teenagers are merely potential lives and can be destroyed".   Why would inherent moral worth depend on the number of neurons or complexity of the brain? 
-5Jakub Supeł

Hey @GeneSmith, really appreciate you putting this together.  I wanted to throw in a few thoughts regarding monogenic disease (the disclaimer/context here is I lead engineering at Orchid, and we've put a lot of thought into our monogenic embryo screening).

The 1% risk of monogenic disease you cite above is pretty misaligned with most estimates of monogenic disease.  It may be an old reference, or filtering for only catastrophic disease, but common estimates for monogenic screening yield in adults are between 3.5% and 18%:

  • BabySeq found 18 out of 159 infants (11.3%) with diagnosable and actionable disease variants.

The BabySeq project revealed unanticipated monogenic disease risks in 11% of newborn babies.

  • DiscovEHR found 3.5% of adults with pathogenic actionable variants

In a survey of 76 clinically actionable disease-associated genes, we estimated that 3.5% of individuals harbor pathogenic or likely pathogenic variants that meet criteria for clinical action. Review of the EHR uncovered findings associated with the monogenic condition in ~65% of pathogenic variant carriers’ medical records.

... (read more)
Thanks a lot for the comment. I'll amend the post with some of this information in the next week. If your numbers are correct (and I have no current reason to doubt them, that substantially increases my estimate of the effectiveness of whole genome embryo sequencing. I've been meaning to write a whole post about the different screening companies but a combination of little time due to starting a new company and a lack of clear data have preventing me from doing so thus far. With this information I might reconsider. One more thing I'd like to ask at some point is whether you're going to publish the AUCs of all the predictors in your panel within some reference population. That would be extremely helpful for patients trying to compare Orchid vs Genomic Prediction or any other company.

You are recommending "Expecting Better" by Emily Oster for prenatal care?

Isn't that the author who claims that pregnant women can have “1 to 2 drinks a week in the first trimester (!)” and one drink daily (!) afterward with no worries?

That is terrifyingly risky advice, contradictory to all major guidelines (CDC, WHO, American Pregnancy Association, the UK Health Ministry, Fetal Alcohol Syndrome advocacy groups...). There have been significant mental changes in babies from quantities as low as that. https://ajp.psychiatryonline.org/doi/epub/10.1176/appi.ajp.2020.20010086 Whether your kid will be one of them is a matter of luck, depending on factors we do not yet understand, but it is really not a bet I would take, and certainly runs counter to the aims of polygenical screening.

I'm a little too tired to read through the study you sent me. I suppose it's possible she either got alcohol wrong or that new studies have come out since the book was published which should change the conclusions. I would note that it's very hard to assess the impacts of drinking in the current environment because women who adhere to health advice about not drinking are probably going to pass on genes to their children that make them less prone to the very neurodevelopmental disorders the studies purport to measure. You could do a study where you find women who you think are at high risk of drinking during pregnancy and then perform an intervention to encourage some of them not to drink. But given that ERBs adhere to a strict Copenhagen interpretation of ethics I doubt such a study design would ever be allowed (even though conducting it would improve children's health if alcohol does indeed have a negative impact).
It stands out to me that she contradicted guidelines by all major agencies and metareviews on this issue, specifically reassuring mothers that it is safe, when so much data points to extreme harms in large quantities, and to the fact that these harmful correlation lower, but stay significant all the way to women who have one single drink. Also that she stuck with it despite having this pointed out to her. It is possible that there is another explanation for the incurable brain damage, and more research is certainly a good idea, but that is a far cry from denying such a highly plausible risk to encourage an behaviour that is completely unnecessary. The fact that she does in this situation has me sceptical of all the rest she recommends, too. I am also not sure whether your correlation idea holds. Many documented cases occur early in the first trimester, because the woman in question is unaware that she has gotten pregnant, and quits as soon as she learns. I doubt that a woman who believes she is not pregnant having a single glass of wine is indicative of anything else you'd expect to be highly correlated with the characteristic fetal brain damage and facial changes we then encounter. The fact that this has become so common that we no longer even recognise those facial changes indicating the brain damage as strange at all is frankly frightening. Importantly, the negative impact of even very little alcohol on the fetus outweighs the IQ benefits you would be spending 20-100 k on. 
FWIW, here's a Cochrane review on RCTs of such encouragement designs. It's basically a failed meta-analysis in that the data is too spotty to make any conclusions. But it shows at the very least that such studies have been done before. An encouragement design in a population with high alcohol use seems most promising for figuring out the causal effect on fetuses, as you say. (Another recent meta-analysis on the same question finds a slight decrease in preterm birth in the alcohol education group, but this is based on just three studies and marginally significant so I'd say it's still uncertain.)
These seem most consistent with the effect of light drinking being small or zero, though it's hard to be certain with the sample sizes given and the difference in study designs. I wish someone would do a large, well-designed, well-run RCT of an encouragement design so we could have a definitive answer to this question.

Great post.  Thank you.  Fertility doctor here and a supporter of ART (assisted reproductive technologies) in general.  A few thoughts (although you touched on a few of these below, worth emphasizing in my opinion):

  • PGT-P has not been validated yet, which may take decades to do, if ever.
  • The science in terms of GWAS isn't quite there yet IMHO - we don't know all the genes that are important for most traits and we may be inadvertently selecting against some desirable traits, for example.
  • Comparing clinic success rates using CDC data is imperfect
... (read more)
I made my reply to your comment into a standalone post
I think you are not taking the eugenics and discrimination concerns seriously enough. I think few people disagree with the idea of selecting the embryo with fewer avoidable diseases that would involve severe suffering; I do not think alzheimers makes humanity better. But there is a slippery slope towards a scenario where people select for sex, skin (and hair and eye) colour, not being queer, not being neurodivergent... and I do find that a dystopian scenario where we would lose something valuable that enriches our world. It makes me worry about and for the people who would opt out of being fixed, or having their children fixed (or rather, discarded early and replaced). I wish I didn't have the genetic tendency towards disease that I unfortunately have, and that causes me a lot of suffering. But I assure you, if my parents had gotten to pick, they would also have made sure that I am not gay. Or tall. Or enby. Or have ADHD and autism. Things that were challenges, to be sure, but that I do not see as a net negative. I also know that both my parents found it disagreeable that I am highly gifted, and smarter than them; my father explicitly considers it a disease state and unpleasant complication, and would certainly have selected against it. I am very glad that he could not. We also already have a scenario where it is incredibly difficult for a poor person to have proper upwards social mobility. If their competitors are literally superior from birth, prior to also getting their personal tutoring and private schools and trust funds and brain implants... at some point, no amount of hard work will make you competitive anymore, the different classes will become insurmountably separated and fixed. Not because the rich parents want this - they just want the best for their children, who wouldn't. But if access to this tech is not fair, and its legitimate usage is not carefully reflected and set, the changes could be dystopian indeed. This tech is transformative. I see how it
Yes, this is actually a fairly common critique of embryo selection. One useful intuition pump I've found helps me think about it is the reversal test; should we make people sicker or more mentally distraught to enrich the world? It's a bit odd to imagine that evolution somehow put us at the perfect equilibrium where any increment or decrement in mental illness rates would result in a worse society. It's especially odd to think that since evolution doesn't care at all about either of those things except insofar as they affect inclusive reproductive fitness. Also, my experience so far just talking to people makes me think parents are going to have different priorities regarding the traits they select for.  I'm sorry about your parents. That sounds like an unpleasant experience. I don't think the thought experiment of "erasing" someone like you from existence is really a very good test of the morality of embryo selection. You are a person with decades of memories and ties to the community of people around you. In my view the morality of "erasing" you feels a lot different than making a choice between two embryos. Unless you believe in souls or something, an embryo is almost pure genetic potential. It has no internal organs, let alone a brain. Even the placenta hasn't formed yet. I would point out that all the dynamics you described are already true to some degree; there are some people born with such extreme genetic disadvantages (through a combination of parentage and bad luck) that there are some paths in life simply closed to them. Of course embryo selection will increase variance, so your point is still well taken. I've spent a fair amount of time thinking about this and the obvious solution here is just to work very hard to make this technology cheaper and better. If we make enough progress on that front then we can just have the government subsidize the technology and give free access to anyone that wants it. Inequality WILL still increase in the meantime,
3Max H
  Not necessarily disagreeing with your main point ("discrimination concerns"), but I want to note that the status quo is currently that an alien god gets to choose those traits. Given the option to wrest (more) control away from natural selection, specific humans, or even humanity collectively, might indeed choose to exercise that control in even worse and more horrifying ways than evolution. So "even more horrifying" is a way things might go, and it's worth weighing that in a cost-benefit analysis. But I just want to note that, as a transhumanist, I regard the status quo, of ceding humanity's collective heritage to an unthinking and unfeeling alien optimization process, to already be pretty horrifying. Partially, I'm horrified at some of the actual outcomes it produces, but I also have a larger philosophical discomfort with trusting something as important as the process by which humans get created to such an alien process, in a universe that (aside from humans themselves) is so unthinking and uncaring and unfair.  
I think I have a different perception of nature, and a much more positive one. With every single life form I see, myself included... I look, and think, your parents survived until they brought children. And their parents survived until they brought children. And their ancestors. And their ancestors. All the way back, for 3 billion years, a single unbroken chain beginning in the first clump of life that hung on, that survived, that hung on, grew, diversified, learned, evolved, a chain of survivors longer than I can conceive. What lives in me now, what lives in all of us, went through this incredible forge, and survived. It has me mesmerised. I look at every strange, random thing, and think, this chose itself, it proved itself, it brought us here.  I look at my body and think, this is a freaking self-assembled nanomachine. I started with two cells, one in the literal nanorange, the other so small it can barely be seen, and that only because it already carried some building blocks for cannibalising later. And they self-assembled into an entire human being, every single step depending on the next, and enterprise that seems insane, and yet here I stand, and this body works. I breathe, I move; my heart cells began beating in the sixth week of pregnancy, and they have never for a single minute stopped since. I can consume practically any organic substance, and my body will disassemble it, and then reassemble it into more of me (how crazy is that?), while breaking down invaders and poisons in the process. I can be cut, and my body will automatically begin self-repairing. I can be invaded by hostile life forms, and have them eliminated. You can destroy a chunk of my brain, the very substrate of my mind, and it will reroute, rebuild, recover. I walk into a forest, and I am overwhelmed by the beauty, the complexity, the balance of it. I look at plants and think, fuck, this thing is literally made of air. It stole the energy from the sun, grabbed some micros from the soil, a

"GeneSmith"... the pun just landed with me. nice.

Curated. This post is a feat of scholarship, well-written, and practical on a high impact topic. Thank you for not just doing the research, but writing it up for everyone else's benefit too. As someone who's personally tried for polygenic screening for IQ, etc., I wish I'd had access to this guide last year.

Sorry I couldn't get it out earlier! I meant to release this in June of last year but the research project into which IVF clinics are best turned out to be quite a bit more difficult than I anticipated.

@GeneSmith, thanks for putting together such an interesting read! Echoing @Ben Podgursky's comments (I’m a Genetic Counselor Orchid), we’re very proud of our monogenic screening that really goes above and beyond carrier screening that’s typically offered to IVF patients. We’re frequently asked “how often do you find something” on this screening. Given that Orchid is the first PGT lab to offer general screening for clinically significant monogenic conditions, there isn’t data available for direct comparison. Instead our detection rate estimate is based on w... (read more)

My partner and I put some effort into benefits from polygenic screening, but alas weren't able to make it work. 

Quick details: we had IVF embryos created and screened for a monogenic disease, (1) this didn't leave us with enough embryos to choose anything, (2) our embryos were created and stored by UCSF clinic, and any screening would have required transferring to another clinic which would have been time consuming and expensive. Unfortunately two rounds of IVF implantation were unsuccessful, so notwithstanding the monogenic disease risk (unclear how ... (read more)

Hi Ruby, Sorry to hear your IVF process didn't work out. UCSF was in the top 59% of clinics nationwide in 2020 and the top 38% in 2019, so while the clinic you chose may not have been the best, you at least didn't pick a bad clinic. Your experience is unfortunately fairly common among IVF patients. Most parents using the procedure are just hoping for at least one child through the process, and many don't have enough embryos to even consider polygenic screening. I really hope someone does a clinical trial of embryo splitting soon. There's a roughly 50% chance of success using the process in animals. I bet with research we could get it up to 80-90%, which would make it viable for increasing live birth rates among parents who don't have many embryos. That's the type of procedure which would have improved the odds of success for parents like yourselves.

I think there is a lot of space in China for a startup to tackle a lot of these problems. What are the exact things that companies in the West don't offer, and how difficult are they to do? I assume that all sequencing would be done locally, and the Chinese side just needs to handle data analysis Westerners are reluctant to do?

One place with a lot of low-hanging fruit is just making predictors that perform better for east asians. There are groups working on this at the moment but none of them have released high quality East Asian predictors to the public yet. Another would be offering screening for traits besides just disease. There may be some companies doing the latter in the US but they are not public at the moment. Another really big differentiator would be if you can create better predictors for non-disease traits. There are many important non-disease traits that parents would likely want to select for, but the predictors for them are not good simply because biobanks don't collect the needed data. You COULD start your own biobank, or perhaps if you have really good connections, get one of the Chinese biobanks to collect the data for you. Among the traits I'm thinking of would be things like: * personality * mental energy * tolerance to sleep deprivation * pro-social tendencies * interpersonal skills Our predictors for these traits are terrible right now. It's worth noting though that there is already at least one Chinese company offering polygenic embryo screening for diabetes. If you wanted to make your own biobank it would probably cost on the order of $20-$50 mil. So pretty expensive for a startup. If you could get access to genotype and phenotype data from Chinese biobanks (or even those from Taiwan or Japan), you could probably train some pretty good predictors for maybe $100k-$200k of labor from a reasonably skilled data scientist. But you'd then run into the problem Orchid is having right now: convincing clinicians to offer your test, or convincing patients to request it. That's actually one of the most underrated challenges. By default, the answer from most clinics will be no. And if you're offering a secondary service, you will need to do a lot of educational outreach because most patients and clinicians have never heard of PGT-P. And those that have are often mi

The LessWrong Review runs every year to select the posts that have most stood the test of time. This post is not yet eligible for review, but will be at the end of 2024. The top fifty or so posts are featured prominently on the site throughout the year. Will this post make the top fifty?

Correlation or causation?

Could you site the studies that this section was based on. I would be interested in reading further as this seems to be the sticking point for most people when it comes to the topic of GM for embryos.

https://www.nature.com/articles/s41598-020-69927-7 This is one of the better papers I know of examining sibling validation. To quote from the article: There's more in the paper if you care to take a look.

I just spoke to someone at Orchid who says their basic pricing is $2,500 per embyro for their most thorough testing, but they do have some sort of volume discount. They didn't tell me when the discount kicks in. 

Yikes. That's even higher than I had heard.

Regarding adding IQ tests to biobank data: I doubt this will happen soon. Steve Hsu recently indicated in a podcast with Alex Murshak that he thinks it will still take many years before biobanks will have a sufficient mass of IQ tests to go along with their data. Of course, this isn't due to a technical hurdle. (The Wonderlic, for instance, is an easy test you could administer to biobank participants at low cost.) Rather, it's due to the fact that this research is in general verboten in the West; in China, the failure to gather IQ test data along with ... (read more)


I have a friend who's a geneticist and is much more pessimistic about the expected impact of embryo selection on a given desired trait. I chatted with him and read some papers he suggested and I now have a lower estimate of the expected value of embryo selection than before.

Very curious to hear from other experts in this field who disagree, especially those who don't have a financial conflict of interest!

It seems it's quite unclear how to even measure how heritable a given trait is, with different methods giving wildly different answers.  And even if ... (read more)

Yes, I hear criticisms like those made by your friend all the time. They aren't particularly discouraging to me. The variation is not that much lower. The standard deviation of any continuous trait is about 70% that of the general population. That's still plenty for selection. The impact from rare genetic variants is mostly NOT captured in today's genetic predictors, and that's a place where the field could improve more in the future. But you don't NEED rare genetic variants for embryo selection to work. You just need a predictor that correlates strongly enough with the actual trait for selection. And for many traits we already have that. The paper you linked showing heritability is lower using relatedness disequilibrium regression was interesting. They don't include intelligence or most of the disease predictors on which embryos are currently selected in the table, so it's hard to give a good estimate of how strongly we should expect it to affect future increases in the efficacy of embryo selection. If I just kind of blindly extrapolate based on the heritability of height in their study and what I know about the heritability of height vs intelligence in other studies, I conclude that the RDR method would generate an estimate of the heritability of intelligence of about 35-40%, which sounds pretty low, but not absurd. It seems pretty likely we could still get a predictor that explained 30% of variance in the trait just using SNP arrays and large sample sizes, so my estimate of the efficacy of embryo selection for intelligence in the future doesn't change much based on that. This is true, but from everything I've read this seems to not matter that much for one simple reason: rare alleles with small effects don't account for that much of the variance in trait values. Like if the RDR method's estimate of heritability is accurate for height, we should expect rare variants to raise the variance explained from 45% to 55%, which would result in a ~10% increase in exp

each disease is weighted according to its impact on disability-adjusted lifespan

It's a pity they don't use some more accurate well-being metrics like f.e. WELLBY (although I think WELLBY isn't ideal either).

How much control do the parents have on what metric will be used to rank the embryos?

I know Genomic Prediction at least has used "Quality Adjusted Lifespan" in their past papers, so I think they're used interchangeably. Both they and Orchid provide the expected absolute lifetime risk of each disease in their reports, so parents can re-prioritize embryo implantation according to their preferences. Doing this methodically is pretty tricky though; every disease has a different age of onset distribution and a different impact on life expectancy and quality of life. My hope is that one or both of them create more granular tools to help patients pick embryos according to their values and preferences. But so far the best you can do is just looking at the raw numbers and googling stuff about average age of onset etc.
3Filip Sondej
Oh, so the option to choose all of those disease weights is there, it's just a lot of effort for the parents? That's good to know. Yeah, ideally it shouldn't need to be done by each parents separately, but rather there should be existing analyses ready. And even if those orgs don't provide a satisfactory analyses themselves, they could be done independently. F.e. collaborating on that with Happier Lives Institute could work well, as they have some similar expertise.
No, the option to select against all diseases in proportion to their impact on quality-adjusted lifespan is the default. But parents can re-do the calculation to like take age of onset into account if they want. Or they could add other non-disease traits to their selection criteria (like intelligence, as estimated by some third party service). I agree, it's very sub-optimal for parents to have to do all this themselves.

A great overview, thanks for writing it! I think it will help a lot of people. I suggest a few corrections though:

That unbelievable, super exponential drop was made possible by a technology called "Next Generation Sequencing".

By the mid-2010s, you could sequence all the parts of a person’s genome most likely to differ from other people’s for under $100.

SNP genotyping doesn't use next generation sequencing, it uses microarrays. (Also it's more properly called genotyping not sequencing)

This is very fortunate for us because these cells contain a treasure trov

... (read more)
Thanks for the correction
Sorry I may have deleted your response to my mosaicism comment when I merged it with this one. I agree that it's controversial.

This is very fortunate for us because these cells contain a treasure trove of information. The most common thing IVF clinics look for is aneuploidy, which is a medical term meaning "this embryo has an abnormal number of chromosomes". The term for this type of testing is “PGT-A”, and it’s performed in roughly half of all IVF cycles in the US today.

Human embryos with the wrong number of chromosomes are surprisingly common, both among IVF patients and natural births. But this wasn’t very well understood before the first use of pre-implantation genetic testing

... (read more)
Yes, I thought about mentioning mosaicism but it's such a can of worms with so much debate in the PGT research field that I just left it out. My reading of the research suggests that mosaicism is far less common than the average PGT test would suggest. Here's a well done study from 2021 in which the authors took apart 942 embryos and found the true rate of mosaicism to be 3%. Most PGT labs diagnose mosaicism at far higher rates: 5-15% at most labs. My impression is this is mostly an artefact of noisy, low-density NGS sequencing technology and/or contaminated samples. It's also worth noting that we have a single digit number of examples of "mosaicism" in actual people, which suggests to me that there's either differential apoptosis of aneuploid cells during embryonic development, or that mosaicism is almost always lethal.

I’ve been wanting a LW article on polygenic screening for some time now, so thank you so much for writing it. I’ll be going through it in detail later. For now, I’d just suggest rewording this:

“The average gene has a tiny, tiny impact on how long you spend in school. The predictor used 2,925 genes to explain just 15% of the variance in how many years of school a person completed.”

Instead of talking about the average gene, consider pointing out that “the most important gene” has only a tiny tiny impact on how long you spend in school. Of course, with 20-25,... (read more)

Thanks for the suggestion. I've edited the post.
It may not have saved, it still reads the same way to me (on a different device, so it's not just cached or anything like that).

Another crucial insight from these studies is that nearly all of the genetic differences between humans can be explained by additive effects; there are very few gene-gene interactions going on; If gene A makes you taller, it doesn't depend on gene B being present to work its magic. It's a strong, independent gene that don't need no help.

This fact is extremely important because it makes both evolution and embryo selection possible. There is a common misconception that genes are tied together in a hopelessly complex web and that if we mess with one part of

... (read more)
There's about 4-5 million letters in the genome where at least one percent of humans have a different letter at that location. That's compared to 3 billion letters overall. Another way to look at genetic differences is to pick a random pair of humans and ask how much they are likely to differ. The answer is by about 3 million base pairs. That's not my impression from reading the literature. There was some giant analysis of educational attainment done last year which found literally zero gene-gene interactions. But I'm not a deep expert on this subject. For intelligence? You can probably get to 1/3rd of variance explained just using SNP arrays like they collect for 23&Me. With whole genome sequencing and more samples you could probably get up to 45%, maybe higher. Gwern has written quite extensively about this. This is not a theoretical assertion but an empirical one. We have studies on educational attainment with like 3 million participants now that have shown ZERO gene-gene interactions. They definitely exist, (at least for other traits) but according to the authors I guess you need an even larger sample size to identify them. Given how little they expect to improve the predictors power by increasing the sample size, one can infer that these interactions, if they exist (and they surely do to some extent), just don't explain very much of the variance. (Ctrl+F for "epistatic interactions" in this paper)
Ok. I guess that, for two random humans, you expect almost all 20000 genes to differ at least on a letter, right? Ok, but this shows that your models do not see the non-additive effects, not that there aren't any. I don't know exactly how analyses are done, but assuming they look at interactions with a model like y=β0+β1x1+β2x2+β12x1x2, then they would not pick up the α term in my example because of the hash (the "hash" stands for any very granular and nonlinear function). But actually I think that it would be very weird to have such "stenographic" interactions only, without also simpler ones, so I'm satisfied with your answer.
Many of the differences between human genomes are actually in "promoter" regions. For a gene to be synthesized into a protein a little enzyme has to come over and bind to a spot next to the gene and transcribe the sequence into mRNA. Other differences are in regions that don't seem to affect traits at all. There's a lot of leftover DNA in our genomes from endoviruses, transposons and other events in our evolutionary history. Sometimes the DNA in those regions randomly mutates into something useful and evolution will start acting on it.

By the mid-2010s, you could genotype all the parts of a person’s DNA most likely to differ from other people’s for under $100. At that price point, it became possible to gather genomes from hundreds of thousands of people and assemble them into giant databases that researchers could access. 

I think there's a good chance that transposon count in the DNA differs between people, shotgun sequencing can't tell you how many times a given sequence appears.  

There's a tradition in biology to consider things that can't be easily measured as irrelevant, but that doesn't mean it's true.

The effect of structural variants like that would be bounded by the difference between SNP heritability and full heritability. That's an easy measurement. (And if it was really responsible for much variance, then it ought to show up as a variance component with whole-genomes from long-read sequencing, I would think.) What evidence is there that transposon counts really matter much in terms of total variance phenome-wide?
Transposons activity is not downregulated in the placenta while being downregulated in most other cells. If there are too many transposons in the DNA that likely makes the placenta fail during the pregnancy. As a result, the amount of variance in transposon counts you see when sequencing born people is more limited than that of embryos. If you do IVF you care about the pregnancy not terminating after three or four months.  The evidence is more of a theoretical argument. In the selfish gene frame, transposons (even when they aren't technically genes) "benefit" from copying themselves even when it reduces the total fitness of the organism. You need a mechanism that prevents people with too many transposons from procreating to account for transposons counts not just growing indefinitely.   In the absence of a process that regulates a quantity, you have variance. Lack of observed variance in sequenced adult genomes is evidence of a mechanism for regulation existing. 
To the extent that transposon sequences affect traits and risks but can't be measured we should expect that be reflected in "missing heritability". You may very well be right that highly repetitive sequences like transposons do affect traits, but that's accounted for.

EDIT: An earlier draft of the post failed to account for the cost of medications in the price of IVF and subsequently understated the expected cost of IVF by $4000. The costs table has been updated to reflect the changes.

Given that this is still fairly underground with all the secret startups, contact Jonathan etc, how should this impact on deciding to do this sooner rather than later? From reading this,  it seems like this is not particularly difficult to do in 2023, providing you know where to look, and will likely get easier in the next few years.

My concern is that as this becomes more widespread, sociopolitical pushback will cause it to become more difficult/costly/less effective, either through social pressure or regulation. You have mentioned already that compan... (read more)

I actually see things going the other way: about 30% of Americans already support gene editing for intelligence enhancement, and almost 50% support embyro selection. My prediction is this tech will gain widespread acceptance. IVF was pretty unpopular when it first came out, but over time as people came to see that the children born via the process were normal, it became much more acceptable to the point where nowadays it's kind of weird to view IVF as immoral. Embryo selection for health, intelligence and happiness is completely consistent with our existing values. And it's not scary in the way eugenics is because you don't need top-down government control of reproduction for it to work. If we just ignore AI for the moment (or assume progress slows down a lot due to regulations/restrictions), I predict the fight is going to shift towards cost and access. People are going to be upset that rich people are securing an advantage for their children that they can't afford. But regardless of whether it becomes popular or causes a backlash, it makes sense to freeze embryos NOW. Female age is the single biggest predictor of gain for a fixed price, so the sooner you go through the process the better.

I think you should mention that traits tend to be correlated. Genomic Prediction has a company policy of not telling you the polygenic scores for height, education attainment, or even cosmetic traits like the eye color of the embryo (this last one can be nearly 100% accurate). Understandably, they don't want to be accused of being an eugenics shop. However, there's strong correlation between height, education attainment, and lower mental disease risk. If you optimize for lower disease risk and height, you will get high education attainment as a by product.... (read more)

Hm. I wonder if there are people who simultaneously say that genes don't affect behavior, and also say that genetic screening is bad. Seems like a near-contradiction, but I suspect there are such people.
There seem to be some disease genes correlated with higher IQ. There's speculation about whether genetic conditions in Ashkenazi Jews cause higher intelligence, but there's also a gene that causes blindness in middle age that also appears to raise intelligence by enhancing neuronal signaling. In general, selective breeding of animals for various traits have often managed to produce animals that excel in that trait but are noticeably less healthy overall. At this point, I don't think actually know which genes are tradeoffs and which are just flaws - including genes that trade off against something that doesn't matter as much in the modern world, such as resistance to famine conditions or infections treatable by antibiotics. We certainly can try to weigh the odds and try to avoid the cases where things have Obviously Gone Wrong, but I reserve the right to worry about removing genes from the human gene pool whose function we don't understand.
It's very reasonable to worry about these things, but let me give you a couple of reasons why I am not very concerned about negative plieotropy: 1. We can observe in actual real humans right now that there is a negative correlation between IQ and disease risk. As you have pointed out, there are some exceptions to this such as autism risk. But the general picture is pretty optimistic. 2. Imagine for a moment that we have two sets of alleles: one set increases IQ. The other set increases disease risk. The sets are not disjoint; that is to say there are some alleles which both increase IQ AND disease risk. If we simply select for IQ, we will select both from the set of genes that ONLY increase IQ with no effect on disease, AND for the set of genes that both increase IQ and disease risk. If you add lower disease risk to your selection index, you will disproportionately select for alleles that are in the subset of variants that increase IQ without increasing disease risk. You'll also select for some alleles that have no effect on IQ but decrease disease risk. Now you might be worried that perhaps there's some disease that isn't in our selection index but nonetheless has variants which increase both risk and IQ. Fortunately, there's a way to deal with this too: add traits that are affected by pretty much all diseases to your selection criteria: for example: 1. Of cou 2. Likelihood of developing chronic pain 3. Self-rated quality of life 4. Life expectancy The two companies that currently offer polygenic embryo screening are already selecting against or for some of these to a limited degree by using disease risk as a proxy. But they could and should select on them directly to deal with the problem you described. So I agree that if you select very very hard on a small set of traits, you are likely to get unforeseen downsides somewhere down the road. But two things worth considering here: 1. Some of the health problems in domesticated anima
I actually do agree with you here. At least in the short run, human knowledge probably already is better than past natural selection at picking genes that result in humans that end up healthy, intelligent, and fertile in the modern world that includes C-sections, baby formula, abundant calories, and antibiotics. My wife's family in particular tends to be very unhealthy and die young. She's 31 now and if she manages to recover from her life-threatening medical conditions to the point that she can safely carry a pregnancy to term, it would help very much if we could do something to reduce our future child's risk of obesity, type 2 diabetes, kidney disease, and heart disease so they won't die before the age of 50.
I'm sorry to hear about your wife's condition. I've had family members that have died young from diseases that most people get in old age as well. It's a very painful experience. If your outlook improves and you decide to do polygenic embryo screening, reach out to me and I can help give you a more personalized assessment of your prospects. The size of the absolute risk reduction you can get from embryo selection will actually be higher if your or your wife's risk is higher. So the benefit will likely be higher for the two of you. Also, the predictors for type 2, obesity, and heart disease are already very good, so if those are your concerns you're in luck.
You're right. I'll edit the original post. I originally didn't include it because i couldn't quantify it that well, but it's worth at least mentioning. Again, if you read the post I believe there are probably groups doing this right now. If you're interested in using the service right now you can talk to Jonathan Anomaly as he seems to know how to get in contact with them.

It's probably morally imperative that all parents who have the financial means to do this do it.

I think it’s morally important that we make this choice increasingly accessible, and fight any bigotry against children born with this method and bigotry against their parents. It would take a pretty niche moral stance and cost benefit analysis to make this morally imperative.
I suspect bigotry against children born this way would not work, just because they would be impossible to identify. (Presumably most of them would not even know themselves). Although a future world where someone says: "phwa! You are only smarter and hotter because of gross polygenic screening your parents cheated into you." Reply: "But, I am a lonely child selected from 4 embryos, you have 4 less successful siblings, so you are more selected than I am."
Yes, I might write later about how to make this cheaper. Though what I would write may have limited relevance, as I expect gene editing methods to surpass simple embryo selection within the next ten years.
I'm not quite sure I would agree with this yet, though I can see the case being made for it. I think it mostly comes down to how much you think you can improve worldwide outcomes by increasing the abilities of those at the top vs bringing up those with the least. Iodine supplementation in the developing world, for example, is probably the single most cost-effective way of increasing average IQ per capita worldwide. It also helps prevent other problems like hypothyroidism. So if just increasing IQ per capita is your goal, polygenic embryo selection is not going to come anywhere close to iodine supplementation. Of course, iodine supplementation is not going to give you any more geniuses, and geniuses per capita has an incredibly strong impact on human progress. I really, really wish we could just ban AI improvements and focus on enhancing human intelligence and morality for a few decades. The reason I originally became interested in embryo selection was that I thought that genetic engineering might be a potential solution to the alignment problem (not to mention many of the other problems the human species faces). But it's going to take at least 20 years to work (and realistically more like 30-40) to have a large impact. I'd put the odds of us getting to AGI before that at like 90%. The only path I can see now involves a worldwide ban on AI capabilities improvements.
Suppose a family values the positive effects that screening would have on their child at $30,000, but in their area, it would cost them $50,000. Them paying for it anyway would be like "donating" $20,000 towards the moral imperative that you propose. But would that really be the best counterfactual use of the money? E.g. donating it instead to the Against Malaria Foundation would save 4-5 lives in expectation.[1] Maybe it would be worth it at $10,000? $5,000? Although, this doesn't take into account the idea that an additional person doing polygenic screening would increase its acceptance in the public, incentivizing companies to innovate and drive the price down. So maybe the knock-on effects would make it worth it. 1. ^ Okay, I've heard that this scale of donations to short-termist charities is actually a lot more complicated than that, but this is just an example.
I mostly agree with this perspective with regards to the "moral imperative". But apart from that, it seems to me that a good case can be made if we use personal health spending as a reference class. Even if we only consider currently achievable DALY gains, it is quite notable that we have a method to gain several healthy life-years for a price of maybe $20,000/healthy year (and actually these gains should even be heritable themselves!). I do not know the numbers for common health interventions, but this should already be somewhat comparable. update: Quick estimate: US per capita health spending in 2019 was $11,582 according to CDC. If the US health spending doubles life expectancy compared to having no health system, this is comparable to $20,000/healthy year.
This may be an overstatement. I think the moral minimum looks more like "bring children into the world in a way that's consistent with the value system you plan to teach them and hope that they live by". If you want to teach a value system of global optimization, where every dollar should be spent to have the maximum possible impact to global quality of life, you're probably adopting rather than conceiving anyways... but this great an investment into a single individual is likely inconsistent with those values. If you want to teach a value system of local optimization, where every person ought to first do what's best for themself and their loved ones before attempting to intervene in the lives of strangers, then it might be inconsistent to gamble with a family member's lifetime wellbeing when you could instead have stacked the odds in their favor.
I think this is incorrect. Behaviors and talents are to a large degree heritable. If you want future people to share your values, one of the best ways to do that is to have kids who are disproportionately likely to share those values. And while you can of course attempt to teach your kids your values, genetics plays a major role in determining what kinds of values we adopt. This is one of the major reasons why "not having kids because of climate change" is not just ineffective but actively counterproductive; it ensures there are less people in the future that will be willing to make large sacrifices for the good of the whole.
I agree that many behaviors are heritable, but I model that inheritance as emerging from the intersection of genetic and environmental factors. I hadn't previously considered generalizing from genetic behavioral proclivities to what values people hold. Could you point me toward the data from which you've drawn this conclusion? I imagine that there are enough adoptee studies in the world to point at a link pretty conclusively if one exists, but I'd also like to skip straight to the most applicable ones if you could recommend them.
Political attitudes seem to be about 30-70% heritable. Interestingly, people's genetics seem to have a stronger effect on their attitudes the more politically engaged they are. https://www.nytimes.com/2022/06/01/opinion/politics-genetics-research.html
The improvements in the mental and physical well-being of the poorest of us if this huge sum of money was instead used to give them healthcare, safety from preventable, transmissible diseases (like malaria nets), clean water, healthy food, crucial supplements, reproductive rights, and most of all education, vastly, vastly outweighs raises the IQs of rich children by this little.
"Rich people" already give the lions share to anti malaria charities, just as virtually all of Earth's economic surplus (for now, pre-AGI) comes from fairly-high-IQ people doing functionally prosocial things. The question is not "is standard EA behavior better than good embryo selection" - effective altruism exists because there are enough altruistic intelligent people around to be EAs - but how good existing methods are, and what runway we have to use them.
I think this is probably true at the moment given the current efficacy and the prices of IVF. The only reason I hesitate is because I think the tendencies and aptitudes of people at the top have a gigantic impact on the rest of society and that improving those tendencies would have a very large impact on the world (especially if parents of children likely to end up in positions of power select for traits like kindness and pro-social tendencies). But if you ignore that for a moment, then you're of course correct that embryo selection can come nowhere close to the efficacy of distributing bed nets or supplementing iodine-deficient populations. However, it's going to take some time to bring down the price of embryo selection and future technologies for genetic improvement, so I think it's probably good to start on that now. Granted, this is all kind of ignoring the 800 pound gorilla in the room which is AI. For embryo selection to really matter, there needs to be time for these children to grow up. If we make ASI in the next 20 years, there's probably not much point. I'm still hopeful that we can get a pause in place on AI development. But without one this whole endeavor has pretty low odds of achieving anything IMO.
"Moral imperatives" is not a category that relies upon financial means. Moral imperatives in traditional Kantian framework are supposed to be universal, no? Just because some action could be personally and socially very beneficial doesn't make it morally compulsory. The benefits would have to be weighed against opportunity cost, uncertainty, game theoretic considerations, and possible contrary moral systems being correct.
Also, please keep the hell out of the reproductive rights of other people. At the point where you prescribe this (and according to whose ideal standards, exactly?), I genuinely no longer understand the difference that is being claimed to outright eugenics. How long until it is de facto mandatory? 

Well this certainly opens a window I hadn't really thought about before. Very intriguing stuff (have only had time to read over the "Concrete Advice" section, but it was a great overview). Not sure I would totally utilize this at the moment (the cost being a major factor), but I have no doubt this kind of thing will get more prevalent and less expensive as time goes on.

I'm wondering if others are having a bit of cognitive dissonance like myself with the taint of eugenics lingering in the air? This quote

"This is an unfortunate side-effect of the fact that t

... (read more)
Yes, I agree that cost and disparities in predictor strength between genetic ancestry groups are probably the biggest issues with polygenic screening at the moment. I didn't mention this in the post, but the issue with disease predictor disparity should improve significantly in the next year or two. There are two large biobanks coming online with significant numbers of non-European participants; The Taiwan Biobank and the "All of Us" biobank in the USA. These should at least reduce the disparity in disease predictor strength, though it remains to be seen whether either one will actually collect data on cognitive phenotypes. I think we need a new term to describe "genetic improvement" that includes embryo selection and people choosing who to reproduce with but not state-sponsored sterilizations or genocide. The fact we use one word to describe both of those is crazy. It's like using "stuff" to refer to both food and excrement. "Oh yeah the stuff I got at the restaurant was good, but stuff isn't always good so eating there made me a bit nervous" Aella girl has suggested the term "epilogenics". "epilogi" is the greek word for "choice", so we can use that instead of "eu" which means "good". I quite like this term and will be using it from now on. So yes, I strongly oppose many types of eugenics but I am strongly in favor of epilogenics.
Yeah I tried to invoke the notion I had that there was this almost ephemeral, tenuous connection between this "stuff" and eugenics. Not trying to imply a direct line of similarity by any means. And I agree that new terminology is needed to distinguish and make useful which aspects society is willing and able to allow vs. not. I'm glad more biobanks are coming online! I could imagine this increasing by an OOM over the next 5-10 years?

Do you have a source for the misconception being common? 

No. This is drawn from my own personal experience reading comments by otherwise knowledgeable academics or professionals in the ART industry. It's also something I read online a lot in forums such as reddit. Inevitably one of the top comments on most articles about human genetic engineering is along the lines of "we don't understand anywhere close to enough about genetics to make the proposed changes". If this is really important to you I can probably find some examples.

you will not be able to have a baby Einstein unless you are already a Nobel laureate...

patently false. 

Neither of Einstein's parents were Nobel laureates; 
his father was a salesman, his mother was a pianist (among other talents).

This is not meant as a literal statement. I'm just trying to convey that the magnitude of benefits is still rather small. There are obviously many examples of geniuses who didn't come from unusually impressive families.

You can see twin birth rates fell sharply in the late 90s

Shouldn't this be triplet birthrates? Twin birthrates look pretty stable in comparison.

Human embryos with the wrong number of chromosomes are surprisingly common, both among IVF patients and natural births.

Do you mean "natural pregnancy?"  Aneuploidy generally results in miscarriage, which we don't call a "natural birth."

Yes, thanks for the correction

Thanks for this awesome post. Biggest update for me is "there might be a way to get screening for traits not advertised by Genomic Prediction", but I still have no idea of the cost or the probability of success :-) Would be great to hear from people who have more info on this.

I recently weighed the pros and cons of IVF vs old-fashioned conception and went old-fashioned because:

1. This article claims "different embryo culture media give rise to different birthweights and growth patterns in children" and "children born after ART have altered epigenetic profi... (read more)

I wrote the section on cost to give you a better idea of the prices involved. Hopefully that's helpful. But I take your point that what is really needed is a “calculator” of some sort where you can input relevant variables and see your expected gains and costs. I am working on something like this at the moment but it may be several months until it's finished. Apart from the randomized control trial looking at different embryo media, I find all the studies presented in this paper to be highly suspect. For example they cite showing that fresh embryo transfer is associate with preterm birth. But the study THEY cite doesn’t even control for the differences in maternal age between parents that do IVF and those that don’t! And surprise surprise, there is a major difference in PTB rates between women in their late 20s and early 30s and those in their late 30s to early 40s. Perhaps I am wrong about this, but my best guess right now is that the downsides of doing IVF are very minor and are massively outweighed by the upsides of embryo selection. The cost is still a big barrier, so I can understand if you don't do it for that reason. Yes, this is still a problem. For IQ gain in particular though the difference is not that big. I believe east and south asians, for example, have an expected IQ gain of about 75% that of Europeans (so like 3.5-4 points vs 5). Maybe that's a big enough difference for it to not be worth it, but it's not a huge reduction.

One thing I didn't see on perusal was a reference to the length of time some of these paths might take. 

I know this could vary extremely according to each person's desired timeline w/r/t keeping embryos frozen and determining the right time for implantation, etc.

But from a naive standpoint, saying that someone wanted to have children quite soon otherwise, how long might one expect to spend going through the "average" or "optimal" path doing this?

There's generally a waiting time of 1-3 months for initial consultations, and another 1-2 months of checkups, hormone treatments and transfers. So 2-5 months before you can actually transfer an embryo, and another 9 months until birth.

Thank you!

Hi, you suggested freezing embryos instead of eggs. However, a recent study found that frozen embryo babies are 2.5 times more likely to develop cancer. At the same time, a review article on IVF comes to the following conclusion (likely due to epigenetic factors):

IVF is associated with an increased health risk for the offspring in the form of malformations, functional dis- orders, and a poorer peripartum outcome. This is possibly caused by parental factors, but also by factors related to the IVF technology used. [...] Consequently, IVF should only be perfo

... (read more)
Thank you for leaving such a thought-provoking comment. I've spent a couple hours reading through the study you posted tonight as well as others linked to by the authors. I don't see the claim about a 2.5x increased risk of cancer anywhere though. From the findings section: So the risk of cancer was 8% higher in those born after ART, and 59% higher for frozen embryos vs fresh embryos. I think the generally higher disease prevalence among IVF couples probably explains the 8% increase for ART in general, though the 59% increase they see for frozen embryo transfer is surprising. Looking more into the study it looks like about a quarter of the effect is driven by the higher rates of twin births in IVF, which are much less common nowadays. This study uses data that is also quite old; they include cycles going all the way back to 1984 or 1994 for some countries. The rate of embryo freezing at that time were quite low, as evidence by the huge difference between hazard ratios for all ART and frozen embryo transfer. If frozen embryos made up a higher proportion of the births you would see a smaller difference between all ART relative to spontaneous conception and frozen embryo transfer relative to spontaneous conception. Here's another study that found higher risk of neoplasms among embryos that were transferred fresh. Granted, this was a smaller study, so I'd lean towards believing your study. There's also a graph in the study which seems to show the relative cancer risk for frozen embryos declining over time: Though this could just reflect fewer twin births. And the confidence interals are such that it's hard to be certain the effect is real. Another possible confounder here is maternal age. The average age of mothers in ART were older than those in the spontaneous conception  group by about 4 years. You can see in this study that maternal age is significantly associated with higher childhood cancer risk. So this could explain another 5-10%, particularly if the

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.
[+][comment deleted]10