I’m not sure I understand what is meant to be communicated by the graphs at the end, but that’s mostly due to my lack of familiarity with these types of formulations. Is the idea that normal measurement error biases our estimates of the genetic correlation of phenotypes upwards?
As far as the issues with twin studies go, I think a far better and more scalable solution is the sibling GWAS study. Siblings differ in their genes, but are raised in a substantially shared environment. There are tons of siblings in the world, so it is much easier to get hundreds of thousands or someday even millions for studies. With samples that large, it is relatively easy to determine which SNPs influence any of a list of traits.
It’s true that environmental change over time can sometimes break the connection between genotype and phenotype, which is why I think we need more cross-generational validations of GWAS predictors. But maybe I’m missing your point?
Is the idea that normal measurement error biases our estimates of the genetic correlation of phenotypes upwards?
The sizes of the genetic correlations in my example are fine; the simulation has the phenotypic causal effect be 0.7, which matches the genetic correlation.
The issue with genetic correlations are more that people might interpret them as referring to horizontal pleiotropy or population structure (narrowly genetic/biological phenomena) when really they can arise from all sorts of phenotypic links between the variables.
And this issue then gets intensified by measurement error because measurement error biases the environmental correlation downwards, making it look like we are dealing with a narrowly biological thing.
As far as the issues with twin studies go, I think a far better and more scalable solution is the sibling GWAS study. Siblings differ in their genes, but are raised in a substantially shared environment. There are tons of siblings in the world, so it is much easier to get hundreds of thousands or someday even millions for studies. With samples that large, it is relatively easy to determine which SNPs influence any of a list of traits.
Within-family genomics is great and I would love to see more of it. However, the basic points in my post still apply. E.g. they’ve found a whole bunch of SNPs that are relevant for educational attainment.
It’s true that environmental change over time can sometimes break the connection between genotype and phenotype, which is why I think we need more cross-generational validations of GWAS predictors. But maybe I’m missing your point?
The point is that I regularly see people treat heritability + genetic correlations as an alternative hypothesis to phenotypic causality, and that this is a misunderstanding of what heritability/genetic correlations tell you.
I’m not sure I understand what is meant to be communicated by the graphs at the end, but that’s mostly due to my lack of familiarity with these types of formulations. Is the idea that normal measurement error biases our estimates of the genetic correlation of phenotypes upwards?
As far as the issues with twin studies go, I think a far better and more scalable solution is the sibling GWAS study. Siblings differ in their genes, but are raised in a substantially shared environment. There are tons of siblings in the world, so it is much easier to get hundreds of thousands or someday even millions for studies. With samples that large, it is relatively easy to determine which SNPs influence any of a list of traits.
It’s true that environmental change over time can sometimes break the connection between genotype and phenotype, which is why I think we need more cross-generational validations of GWAS predictors. But maybe I’m missing your point?
The sizes of the genetic correlations in my example are fine; the simulation has the phenotypic causal effect be 0.7, which matches the genetic correlation.
The issue with genetic correlations are more that people might interpret them as referring to horizontal pleiotropy or population structure (narrowly genetic/biological phenomena) when really they can arise from all sorts of phenotypic links between the variables.
And this issue then gets intensified by measurement error because measurement error biases the environmental correlation downwards, making it look like we are dealing with a narrowly biological thing.
Within-family genomics is great and I would love to see more of it. However, the basic points in my post still apply. E.g. they’ve found a whole bunch of SNPs that are relevant for educational attainment.
The point is that I regularly see people treat heritability + genetic correlations as an alternative hypothesis to phenotypic causality, and that this is a misunderstanding of what heritability/genetic correlations tell you.