Thanks for this, this is a concrete post. So the selection step itself is treated pretty informally. Parents either set hard thresholds or build a weighted spreadsheet and either way we’re collapsing a pretty rich posterior into a single number per embryo, the reports already carry more information than that approach uses. Before touching weights at all, we can drop embryos that are worse on every trait the couple cares about. If an embryo is strictly dominated by another, there’s no preference structure under which it’s the right pick so we can eliminate it geometrically and remove a chunk of “did we pick the wrong weights” regret for free
and when parents do have to weigh the remaining embryos, the usual ‘how many iq points equal one percent of schizophrenia risk’ framing treats each trait as independent and flattens the variance, I find the fritz story is a nice illustration of the alternative mattering in practice: so the t1d posterior is bimodal because of the protective variant not just low on average and an expected-value comparison would have obscured how onesided that call actually was
Do any of the companies expose the variance on individual predictions or only the point estimates? 9% ± 0.5% is a different decision than 9% ± 4% when the alternative embryo is at 7%, and i’d guess most parents would rather take a lower mean embryo with tight variance than a higher-mean one with wide variance if the report actually let them see that, if the variance is in the underlying predictor but not in the report, that seems like a cheap product fix imo
Do any of the companies expose the variance on individual predictions or only the point estimates? 9% ± 0.5% is a different decision than 9% ± 4% when the alternative embryo is at 7%, and i’d guess most parents would rather take a lower mean embryo with tight variance than a higher-mean one with wide variance if the report actually let them see that, if the variance is in the underlying predictor but not in the report, that seems like a cheap product fix imo
Ok, I’ve literally checked with the most knowledgeable people in the world on this, and the answer is that prediction error (i.e. the variance in outcomes), does vary a tiny but, but only as a result of differences in the imputation accuracy of each embryo’s genome.
Herasight is the only company that does embryo genome imputation at the moment, so this isn’t even relevant to Genomic Prediction or Orchid.
Also, thinking about “error” in the prediction only really makes sense for continuous traits like height or IQ. For disease risk, any company worth dealing with account for prediction error in their risk prediction.
A weaker predictor should always produce a disease risk that’s close to the population average for that particular embryo’s sex and genetic ancestry. A stronger one will show more variance in risk from one embryo to the next.
And this is in fact exactly what you see. There’s quite a large spread in the risk of a disease like type 1 diabetes, and quite a small spread for something like epithelial ovarian cancer.
For IQ, the standard deviation of IQ for a randomly chosen embryo would be +-10.6 (15/sqrt(2)). If you know the predicted IQ from Herasight’s IQ predictor, that variance shrinks to 9.3 because a portion of the variance is now “fixed” thanks to the prediction.
And yes, Herasight shows this. Neither Genomic Prediction nor Orchid offer selection on continuous traits right now. I don’t know if Nucleus does, but they have other issues such that confidence intervals should probably be the least of your concerns.
Thanks for this, this is a concrete post. So the selection step itself is treated pretty informally. Parents either set hard thresholds or build a weighted spreadsheet and either way we’re collapsing a pretty rich posterior into a single number per embryo, the reports already carry more information than that approach uses. Before touching weights at all, we can drop embryos that are worse on every trait the couple cares about. If an embryo is strictly dominated by another, there’s no preference structure under which it’s the right pick so we can eliminate it geometrically and remove a chunk of “did we pick the wrong weights” regret for free
and when parents do have to weigh the remaining embryos, the usual ‘how many iq points equal one percent of schizophrenia risk’ framing treats each trait as independent and flattens the variance, I find the fritz story is a nice illustration of the alternative mattering in practice: so the t1d posterior is bimodal because of the protective variant not just low on average and an expected-value comparison would have obscured how onesided that call actually was
Do any of the companies expose the variance on individual predictions or only the point estimates? 9% ± 0.5% is a different decision than 9% ± 4% when the alternative embryo is at 7%, and i’d guess most parents would rather take a lower mean embryo with tight variance than a higher-mean one with wide variance if the report actually let them see that, if the variance is in the underlying predictor but not in the report, that seems like a cheap product fix imo
Ok, I’ve literally checked with the most knowledgeable people in the world on this, and the answer is that prediction error (i.e. the variance in outcomes), does vary a tiny but, but only as a result of differences in the imputation accuracy of each embryo’s genome.
Herasight is the only company that does embryo genome imputation at the moment, so this isn’t even relevant to Genomic Prediction or Orchid.
Also, thinking about “error” in the prediction only really makes sense for continuous traits like height or IQ. For disease risk, any company worth dealing with account for prediction error in their risk prediction.
A weaker predictor should always produce a disease risk that’s close to the population average for that particular embryo’s sex and genetic ancestry. A stronger one will show more variance in risk from one embryo to the next.
And this is in fact exactly what you see. There’s quite a large spread in the risk of a disease like type 1 diabetes, and quite a small spread for something like epithelial ovarian cancer.
For IQ, the standard deviation of IQ for a randomly chosen embryo would be +-10.6 (15/sqrt(2)). If you know the predicted IQ from Herasight’s IQ predictor, that variance shrinks to 9.3 because a portion of the variance is now “fixed” thanks to the prediction.
For example, an embryo with a predicted IQ of +10 would have a narrowed, right shifted distribution compared to an average embryo: https://pbs.twimg.com/media/G6oTYmLbwAI51py?format=jpg&name=medium
And yes, Herasight shows this. Neither Genomic Prediction nor Orchid offer selection on continuous traits right now. I don’t know if Nucleus does, but they have other issues such that confidence intervals should probably be the least of your concerns.