I built this for parents undergoing IVF interested in selecting embryos for cognitive function. I used SAT score (with a standard deviation of 200 points) as an illustrative proxy.
In this tool, we assume that a polygenic predictor of cognitive function is computed for each viable frozen embryo. The embryos are ordered by the output of the polygenic predictor. Then, a parent selects among the top ones for implantation.
Based on your selections, I show the probability distribution of SAT score of your selected embryos as compared to your frozen embryos.
In particular, I provide:
• The average improvement in SAT score attributable to embryo selection.
• How much more or less likely each SAT score is for a selected embryo as compared to a frozen embryo. This is represented by the red dashed line.
• The reduced risk that a selected embryo will have low cognitive function (defined as 400 points below the average SAT score of frozen embryos).
User Inputs
• Number of Frozen Embryos is the number of viable frozen embryos analyzed in pre-implantation genetic testing. This is the population the selected embryos are drawn from and compared to. To estimate how many viable embryos you can expect from IVF, see this tool from Orchid Health.
• Number of Selected Embryos is the number of highest-ranking embryos according to the polygenic predictor considered for implantation. For example, if you have 15 total frozen embryos and want to implant from the top third of predicted cognitive function, your number of selected embryos would be 5.
• Polygenic Predictor Correlation is the predictive power of the polygenic predictor used to select the embryos. It is the correlation between the polygenic predictor and the trait of interest in the population.
Background
A polygenic predictor is a weighted sum of common genetic variants associated with a trait. Amongst publicly available polygenic predictors of cognitive function, the highest correlation I'm aware of is around 0.3 (Becker et al. 2021). Current attempts to develop polygenic predictors are strained by low sample sizes in the tens of thousands and imprecise cognitive performance meausures (e.g. a two-minute questionnaire).
It is projected that if only we had the genotypes and cognitive performance scores (e.g. 23andMe files and SAT scores) of 500k to 1M people, the polygenic predictor correlation could be as high as 0.6 - 0.7.
For context, a correlation of 0.63 for height was obtained using paired heights and genomes of 407,849 individuals (Lello et al. 2018).
Code
The code to generate the probability distribution estimates is available here: embryo.py
Contact
If you have any questions or would like to chat about this, please email me at brobdingrag@gmail.com