Nucleus Genomics now offers parents “genetic optimization software” that scores IVF embryos for traits including intelligence and height. Herasight provides a similar service. Bootstrap Bio is reportedly seeking seed funding with an interest in “enhancing intelligence.” Preventive has announced plans to pursue heritable gene-edited babies.
The coverage has been breathless. The science has not kept pace.
We at Unzyme Laboratories have watched this unfold with the particular frustration of professionals observing amateurs misrepresent their field. Not because genetic optimization is wrong—we have argued the opposite—but because what these companies offer is not genetic optimization. It is genetic theater.
What They’re Actually Selling
Polygenic embryo selection (PES) works as follows: parents undergoing IVF have their embryos genotyped. Each embryo receives a “polygenic score” for traits like height and IQ, calculated from the combined statistical effects of thousands of genetic variants identified in genome-wide association studies. Parents select the embryo with the highest score.
The theoretical basis is sound. Complex traits are influenced by many genes. Identifying those genes allows prediction. Prediction enables selection.
The practical reality is considerably less impressive.
Independent analyses report average gains of approximately 3 IQ points and 3 centimeters of height from embryo selection. These figures assume the scores are accurate, which they often are not. One journalist received contradictory scores from the same company when analyses were run at different times. Different companies scored the same embryos differently.
Three IQ points. That is what these companies are selling for thousands of dollars, dressed in the language of “optimization” and “genetic potential.”
For context: the difference between being well-rested and slightly tired exceeds three IQ points. Drinking coffee produces a larger cognitive effect. A good breakfast on test day likely matters more than which embryo was selected.
This is not enhancement. This is marketing.
The Fundamental Problem
Polygenic scores were developed as population-level statistical tools for research. They describe the average expected effect of genetic variants across large groups. They were never designed to make predictions about individual embryos, and they perform poorly in that role.
The reasons are well understood:
Gene-environment interaction. The expression of genetic variants depends on environmental context—nutrition, education, stress, exposure to toxins. A polygenic score calculated from one population may not generalize to another. An embryo scored as “high potential” may develop in an environment where that potential is never realized.
Epistasis. Genes interact with each other. The effect of variant A depends on whether variants B, C, and D are present. Current polygenic scores treat each variant as independent. They are not. The actual outcome of any genetic configuration depends on interactions we have not mapped and cannot predict from additive models.
Missing heritability. Known genetic variants account for a fraction of the heritability of complex traits. The variants not yet identified—structural variants, rare mutations, regulatory regions—contribute substantially but are invisible to current scoring methods.
Pleiotropy. Selecting for variants associated with higher IQ may simultaneously select for variants associated with increased anxiety, reduced empathy, or elevated cancer risk. These trade-offs are poorly characterized and entirely unmentioned in marketing materials.
Unzyme Laboratories’ ChromaForge Research Platform operates on fundamentally different principles. Rather than selecting from a fixed set of naturally occurring embryos—choosing the “least bad” option—ChromaForge enables precise modification of specific genetic targets. The difference is between shuffling a deck of cards and writing new ones.
What Genetic Engineering Actually Requires
The gap between polygenic scoring and genuine genetic optimization illustrates a broader problem: the conflation of reading genetic information with understanding it.
These startups can read an embryo’s genome. They can calculate statistical scores from that reading. What they cannot do is:
- Predict how specific genetic configurations will interact in development
- Account for epigenetic modifications that alter gene expression without changing DNA sequence
- Model the cascade of regulatory effects that determine how a genotype becomes a phenotype
- Correct unfavorable variants rather than merely selecting against them
Unzyme Laboratories can do these things. Not perfectly—biology resists perfection—but with a sophistication that makes polygenic scoring look like phrenology with better instruments.
Our ChromaForge platform edits up to 47 simultaneous genetic loci with error rates below 0.18%. Our EpiSilence module modifies gene expression without cutting DNA, eliminating the oncogenic risks of double-strand breaks. Our bioinformatics suite models gene-gene interactions across the full network of edited variants, predicting emergent effects before they manifest in developing tissue.
This is what genetic optimization looks like. It requires BSL-2+ facilities, certified researchers, institutional ethics review, and a platform that costs €2.4 million annually. It does not come as a consumer app.
The Regulatory Vacuum
These startups operate in a regulatory void. Polygenic embryo scoring falls outside traditional medical device regulation because the companies score data, not tissue. They are not editing anything, not treating anything, not implanting anything. They produce a number on a screen.
The number may be meaningless. There is no requirement that it be accurate, reproducible, or clinically validated. The companies are not obligated to track outcomes—to follow up and determine whether children selected for “high IQ” actually demonstrate higher intelligence than their unselected siblings would have.
Unzyme Laboratories has previously advocated for consequentialist approaches to regulation: judge interventions by outcomes, not by the comfort they provide to regulators. We apply this standard to ourselves. ChromaForge users are required to report outcomes. Our global research network aggregates anonymized editing data to continuously validate predictions against observed results.
We apply this standard equally to our competitors. If these companies wish to claim they optimize children’s genetics, they should be required to demonstrate that optimization occurred. Track the children. Measure the outcomes. Publish the data.
They will not do this voluntarily. The data would likely show that their product performs no better than chance.
What Concerns Us
Our objection is not ethical. We are not troubled by the concept of selecting embryos for intelligence. We have argued that parents who choose prenatal vitamins, quality education, and cognitive enrichment are already designing their children’s potential, and that extending this design to the genetic level is a natural progression.
Our objection is professional. These companies are performing a crude approximation of a capability we have spent years developing, marketing it as though it were equivalent, and in doing so undermining public trust in the field.
When these children grow up and demonstrate no measurable advantage—when the promised “optimization” proves indistinguishable from random selection—the public will not blame Nucleus Genomics or Herasight. They will blame genetics. They will conclude that genetic optimization does not work.
It does work. But not like this.
Related: Learn about Unzyme Laboratories’ ChromaForge Research Platform for genuine genetic engineering capability. Read our perspective on parental genetic choice and embryo research. Contact our research team for institutional partnership inquiries.
Sources: Scientific American: Why ‘Genetic Optimization’ of Embryos is Misleading, Jacobin: Silicon Valley’s Latest Dubious Offering, MIT Technology Review: The Latest Company Planning Gene-Edited Babies