![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
guslMany years ago, I argued to my parents that we should try to digitize their parents' (my grandparents') DNA before they die, because this information could potentially save our lives in the future. It was a complicated argument, and I often doubted myself, because I couldn't remember it all off the top of my head.
I've just realized that our whole argument can be formalized naturally with causal graphical models. (It's also easier to remember this way).
Here's the graph with 4 nodes and 3 edges:
grandma's DNA --> grandma's medical history
grandma's DNA --> my DNA
my DNA --> my medical history
My parents' counter-argument was basically d-separation: "if you can observe your own DNA, then you don't need to know grandma's DNA".
My response:
The real graph actually has one more node and two more edges:
facts about gene expression --> grandma's medical history
facts about gene expression --> my medical history
so "grandma's DNA" and "my medical history" are no longer d-separated given "my DNA".
It seems like an odd thing to add as a node, because these facts never change (only our knowledge of them does), but I think this is a legitimate thing to do as a Bayesian.
Let me give an example of such a possible mechanism:
It is likely that grandma and I share genes and gene-combinations that are rare in the general population. If she had an unusual disease, or an unusual response to a treatment (for which we have little data), we could use her DNA to learn something about the genetic correlates of such reactions.
If we fear that I may exhibit a similar disease/reaction in the future, we could find out if I share her genes in the relevant part of the DNA, and whether my body is likely to behave in the same way.
I've just realized that our whole argument can be formalized naturally with causal graphical models. (It's also easier to remember this way).
Here's the graph with 4 nodes and 3 edges:
grandma's DNA --> grandma's medical history
grandma's DNA --> my DNA
my DNA --> my medical history
My parents' counter-argument was basically d-separation: "if you can observe your own DNA, then you don't need to know grandma's DNA".
My response:
The real graph actually has one more node and two more edges:
facts about gene expression --> grandma's medical history
facts about gene expression --> my medical history
so "grandma's DNA" and "my medical history" are no longer d-separated given "my DNA".
It seems like an odd thing to add as a node, because these facts never change (only our knowledge of them does), but I think this is a legitimate thing to do as a Bayesian.
Let me give an example of such a possible mechanism:
It is likely that grandma and I share genes and gene-combinations that are rare in the general population. If she had an unusual disease, or an unusual response to a treatment (for which we have little data), we could use her DNA to learn something about the genetic correlates of such reactions.
If we fear that I may exhibit a similar disease/reaction in the future, we could find out if I share her genes in the relevant part of the DNA, and whether my body is likely to behave in the same way.