By: Ralph Porneso. Ralph Porneso is a doctoral researcher at PROMENTA in the University of Oslo and is funded by the EU Horizon Project under grant no. 101073237. His area of research revolves around refining and expanding genetic models for complex traits. All views expressed are those of the authors.
Embryo selection startups promise to give parents the option to choose smarter and healthier children. Professional organizations have started to push back on this narrative, calling it premature and reckless. Among scientists, it has exposed a divide. But in most people’s minds, a question remains – what can it really do?
Genetic testing is a powerful technology. For decades, it has been used as a prenatal screening tool for conditions with serious and debilitating consequences. Chromosomal anomalies like trisomy 21, 18, or 13, in which an individual carries an extra copy of a chromosome, can be reliably detected as early as ten weeks into a pregnancy. Genetic testing can also screen for rare monogenic diseases by scanning the genome for mutations with deleterious effects. Individuals with these mutations develop the disease – e.g. those whose HBB genes contain a specific mutation develop Sickle cell anemia, a debilitating condition characterized by infections and episodes of severe pain lasting days or even weeks. Medical professionals in most countries in Europe offer prenatal screening for these conditions, especially to couples who are at a higher risk of conferring them to their children.
In the US, venture capitalists have begun funding companies that expand the use of genetic testing beyond screening for monogenic diseases. Startups – Genomic Prediction, Nucleus Genomics, Herasight, and Orchid – offer expectant IVF parents the option to choose their “best” children. Armed with genome sequencing technology, these companies score embryos using genetic mutations to predict height, IQ, type 2 diabetes, and other health outcomes. Their promise: the higher the score, the taller, smarter, and healthier this future person will be.
But how does it work?
Startups take the scores from observational studies that link DNA mutations (i.e. variants) to traits. Each variant’s score is small. A height-increasing variant, for example, is in the order of 0.1-1 millimeter. A person may get two copies – one from each parent – so the maximum gain from a single variant is twice its score (0.2-2 millimeter, in this example). They are combined into an aggregate score used to predict an outcome or a trait. The more variants in the aggregate score, the better its prediction.
The power of the technology lies in ranking embryos based on their aggregate scores. The ideal scenario is to have at least five embryos. With fewer, the embryos are too similar and ranking doesn’t work because they share roughly 50% of their genetic material. This makes embryo selection a bad fit for infertile couples and is better suited to those who are able and willing to produce ten or more viable embryos.
In December 2025, the American Society for Reproductive Medicine (ASRM) drew the line and issued a statement stating that “current evidence does not support the predictive accuracy, safety, or clinical value of polygenic embryo screening” and “it risks misleading patients by overstating what polygenic risk scores can reliably determine.” The International Society of Psychiatric Genetics (ISPG) takes a similar position. On their website, they state, “polygenic risk scores (PRS) are currently considered research analyses and are not recommended for clinical use.”
When embryo selection companies started to emerge, scientists promptly challenged their claims, warning of the views the technology implicitly espoused. In academic circles, embryo selection is seen more as science fiction. Or, at best, a kind of blind optimism about its promise and reach. Until it came out that geneticists – colleagues and, for some of us, friends – were behind some of these companies. It split the field and gave embryo selection a degree of legitimacy. Conversations around it were no longer just about the science. Ethics, societal impact, philosophy were interspersed alongside discussions of its technical merits and limitations; a few deserve close attention.
The statistical model used to link DNA sequences to traits assumes their scores are constant. Scientists point out this limits their utility because most change and depend, for instance, on genetic background or ancestry. The majority of studies that derived these scores used European populations. Using European scores on individuals from African, Asian, or mixed ancestries results in noisier predictions – even with more sophisticated models that directly correct for genetic background.
In some cases, the scores change even among individuals from the same ancestry if they come from different historical and environmental contexts. To take a radical example, imagine measuring scores in educational achievement in 16th century women. Applying those scores to predict educational achievement today would result in an aggregate score of… zero. This is how genetic prediction works. Scores derived from people from more than two generations ago are applied to individuals today.
There are other issues. Almost all these variants are in segments of the DNA that currently have no known biological functions. Identifying causal links between variants and outcomes is difficult, even with the help of the most advanced molecular techniques today (e.g. CRISPR). In contrast, variants in monogenic diseases are well characterized. They occur in genes or in regions that directly affect genes and their products. Empirical evidence from tightly controlled experiments back them up, enabling scientists to map how single variants translate into diseases and disorders, with clear causal mechanisms that link them.
Some scientists argue that the lack of a causal link does not mean aggregate scores cannot predict future outcomes and traits. The utility of these scores lies in how well they discriminate individuals who express an outcome versus those who don’t. For example, 7-foot people should have the top height scores. But, in practice, it is not uncommon to find top scorers with average trait levels. Or bottom scorers with high trait levels and vice-versa.
There is little benefit in ranking embryos using scores that predict traits poorly. If we quantify how well they predict traits, it is 10% on average1. This number does not mean the prediction will be wrong 90 out of 100 times. Instead, they reflect how much traits are explained by genetics. If a trait is largely not determined by genetics, then DNA sequences will not effectively predict them – no matter how advanced or sophisticated the mathematical models are.
It’s a bit different for diseases. Scientists do not look at how much variants increase or decrease a disease, but on how likely they are to tip a person’s status from healthy to diseased. This is where genetic prediction shows the most promise, especially for highly heritable diseases that are common in a population. At present, scientists are carefully evaluating the aggregate score for cardiovascular disease, the third leading cause of death worldwide. The goal is to prevent high scorers from developing it through early interventions. Scientists are calling for prospective studies, where high scorers are followed throughout life to validate if early interventions decrease their risk.
Genetic testing in embryo selection can also reduce disease risk. A powerful score for type 2 diabetes can lower it by as much as 15%2. Startup scientists, however, assume that a single round of IVF results in a successful pregnancy. This inflates their estimated reduction by almost three folds. The success rate in reality is only 20-30% on average. Given an optimistic but realistic success rate of 50% – say, for fertile individuals – the risk reduction is still non-trivial at 6%3,4.
But there are practical considerations, too. A single round of IVF costs roughly $20,000 and another $12,500 to $50,000 for scoring and ranking embryos depending on the provider. More importantly, rounds of chemical stimulation take a physical and an emotional toll on couples, especially on women who risk pelvic infection, ectopic pregnancy, and ovarian hyperstimulation resulting in pain and discomfort.
Embryo selection is a personal choice. It is reasonable for parents to consider using it to reduce their children’s risks however small they may be. But genetic testing does not have to be done on embryos. Expectant parents can be tested instead for mutations known to result in genetic conditions. If neither carries a disease-causing mutation, then there is no need to go through multiple IVF rounds to harvest and rank as many embryos as possible. Companies like JScreen test up to 260 genetic conditions, including rare genetic disorders. They also offer screens for mutations in BRCA1 and BRCA2 genes to assess risk for different forms of cancer. Couples with health insurance can take advantage of these tests for as low as $35 (or $285 if paid out-of-pocket).
Parents want the best for their children. It is a natural and admirable human desire. Venture capitalists have taken notice and are applying advances in genetic prediction to monetize it. As some scientists move to industry, others now find themselves tempering expectations and grounding embryo selection in reality – a technology that promises certainty in parenting, built on a science that is most honest about its uncertainty.
References
1 Okbay A, Wu Y, Wang N, Jayashankar H, Bennett M, Nehzati SM, Sidorenko J, Kweon H, Goldman G, Gjorgjieva T, Jiang Y, Hicks B, Tian C, Hinds DA, Ahlskog R, Magnusson PKE, Oskarsson S, Hayward C, Campbell A, Porteous DJ, Freese J, Herd P; 23andMe Research Team; Social Science Genetic Association Consortium; Watson C, Jala J, Conley D, Koellinger PD, Johannesson M, Laibson D, Meyer MN, Lee JJ, Kong A, Yengo L, Cesarini D, Turley P, Visscher PM, Beauchamp JP, Benjamin DJ, Young AI. Polygenic prediction of educational attainment within and between families from genome-wide association analyses in 3 million individuals. Nat Genet. 2022 Apr;54(4):437-449. doi: 10.1038/s41588-022-01016-z. Epub 2022 Mar 31. PMID: 35361970; PMCID: PMC9005349.
2 Moore S, Davidson I, Anomaly J, Li JH, Ahangari M, Moissiy L, Christensen M, Young AS, Stern D, Wolfram T. Development and validation of polygenic scores for within-family prediction of disease risks. 2025.08.06.25333145 Preprint at 10.1101/2025.08.06.25333145 (2025).
3 Klausner L, Revital A, Lencz T, Carmi S. PEStimate: Predicting offspring disease risk after Polygenic Embryo Screening. medRxiv [Preprint]. 2025 Sep 9:2025.09.05.25335168. doi: 10.1101/2025.09.05.25335168. PMID: 40963748; PMCID: PMC12440052.
4 Roura-Monllor JA, Walker Z, Reynolds JM, Rivera-Cruz G, Hershlag A, Altrescu G, Klipstein S, Pereira S, Lázaro-Muñoz G, Carmi S, Lencz T, Lathi RB. Promises and pitfalls of preimplantation genetic testing for polygenic disorders: a narrative review. F&S Reviews, Volume 6, Issue 1, 2025, 100085, ISSN 2666-5719. https://doi.org/10.1016/j.xfnr.2024.100085.
This research was supported by the European Social Science Genetics Network (ESSGN) under the European Union’s Horizon 2020 research and innovation programme’s Marie Skłodowska-Curie grant agreement (ESSGN 101073237). The views expressed in this publication are those of the authors and do not necessarily reflect those of the European Union, MSCA Horizon Europe, or ESSGN. Neither the European Union, nor the granting authority or ESSGN can be held responsible for them.
