Two reasons the score is weaker than the brochure
It was measured between strangers and it is being used between siblings. And it was mostly built on one part of the world's DNA.
Reason one: siblings are not strangers.
Polygenic scores were built by comparing large numbers of unrelated people. Embryo selection does something completely different. It compares the embryos of one couple, who are full siblings. Two things happen when you do that.
The NEJM authors explain both: because every embryo's genome is a mixture of the biologic parents' genomes, there is less variation in polygenic scores among the set of embryos produced by the same two biologic parents than among embryos produced by different pairs of biologic parents
, and a sizeable portion of the predictive power of the polygenic score for persons from different families comes from gene-environment correlation, that is, persons with high polygenic scores are likely to be raised in family environments that promote educational attainment.
That second effect has been measured. A 2018 study in Science found that the score built from the DNA a parent did not pass on still predicts the child's education, at 29.9% of the strength of the DNA that was passed on. Roughly thirty percent of what an education score appears to measure is not in the child's genes at all. It cannot be selected for at the embryo stage, because it is not in the embryo.
A twin study found scores for cognitive traits predicted on average 60% better between families than within them, and that much of the difference disappeared once family income and social position were accounted for.
The largest study of this used 178,076 people with at least one genotyped sibling, from 77,832 sibships across 19 studies. Here is what shrank and what did not.
| Trait | How much the genetic effect shrank in a sibling comparison |
|---|---|
| Number of children | 67% |
| Age at first birth | 52% |
| Depressive symptoms | 50% |
| Educational attainment | 47% |
| Cognitive ability | 22% |
| Ever smoking | 19% |
| Height | 10% |
| C-reactive protein | Minus 9%, meaning the estimate was larger between siblings, not smaller |
| The remaining 17 traits, including body mass index and blood pressure | Little evidence of any difference |
Read the bottom row as carefully as the top one. It would be wrong to tell you that comparing siblings destroys most of the predictive power for heart disease or cholesterol. In this study, it did not. What shrinks most is the power to predict education, cognition and behaviour, which are exactly the things being marketed as trait selection.
The European genetics society draws the practical conclusion: The intrafamilial variability would be much more limited than in the wider population, and therefore the PRS would be unlikely to be useful in determining the choice of one embryo over another, particularly as the number of viable embryos available is typically very small.
Reason two: whose DNA the scores were built on.
Prediction accuracy was measured across 17 traits in a large biobank, using scores derived from European data. Compared with accuracy in people of European ancestry, accuracy was 1.6 times lower in Hispanic or Latino Americans, 1.7 times lower in South Asians, 2.5 times lower in East Asians and 4.9 times lower in people of African ancestry.
The authors state it plainly: African descent populations, which already endure many of the largest health disparities globally, are often predicted marginally better, if at all, compared to random.
Because that is a statement about accuracy, here are the absolute figures beside it. These are the same modelled reductions from the previous page, in percentage points, for two ancestry groups.
| Condition | Absolute reduction, European ancestry | Absolute reduction, African ancestry |
|---|---|---|
| Type 2 diabetes | 5.5 points | 2.6 points |
| Coronary artery disease | 1.1 points | 0.55 points |
| Breast cancer | 1.9 points | 0.91 points |
| Type 1 diabetes | 0.12 points | 0.07 points |
And if the two parents are of different ancestries, the European genetics society says the score cannot be calibrated at all: PRSs themselves should be calculated using their effects from the ethnic group the parents belong to. The estimation of PRSs for children of parents from diverse ethnic origins is not yet possible to determine correctly.
ASRM says the same thing in its own words: The databases used to create algorithms for PGT-P derive from samples that are most often collected from adults of Western European Caucasian origin, thereby potentially limiting the generalizability of the data.
"If we are of different ancestries, how was that handled?"
"Was the accuracy figure you are quoting measured between siblings, or between unrelated people?"