Is the takeaway that knowing the actual relatedness of DZ twins doesn’t contribute much to your estimation of ACE model parameters vs. estimating the relatedness at .5 if you have both MZ and DZ twins in your model?
It;s more that you have more power if there is more variation in coefficient of relatedness
the absolute most powerful design is MZs raised apart (but only if they grow up in unrelated and uncorrelated families)
As the variance in the coefficient of relatedness does down you need more N. The sd of relatedness among DZs is pretty small - range .35-.65
I see! So that’s why we didn’t see dramatically smaller CIs here?
I wonder how to calculate actual relatedness (piH) for each DZ pair from real data? Is it the value of Coefficient of genetic relationship Rjk?
Plink (for example) can calculate pi-hat proper from genotypic data. Depending on how you’re calculating Rjk, it is probably a reasonable approximation.
Are we using actual SNP data to get actual relatedness of the DZ pair?
In real life, you could, but in this script, all the data are simulated, and pi-hat comes from line 53:
piH <- rnorm( Np, mean=.5, sd=.03)
But yes in real life you would estimate with SNPs - GCTA was designed to do this (but other programs can now do this too)
If we have a partial genotyped data set what is the recommended approach? can we use both pihat and estimated values?
interesting question - yes you could
How would it handle the missing data though?
In OpenMx, missing data are handled with full information maximum likelihood. This is true for (a) continuous data, (b) ordered categorical data, and (c) combinations of ordered categorical data and continuous data.
just code it as NA?
You would be imputing the missing pi-hats as the expected value, no?
Unless it’s a definition variable!!
this would be a 2 group solution - one with pihat and one with theoretical
I thought you meant missing data in the outcomes.
That’s correct.
One way to think about it is that the genetic similarity between MZ twins is 1, whereas the average genetic similarity between DZ twins is 0.5. That very large contrast in genetic sharing allows you to estimate V_A relatively precisely. In contrast, there is very little variation in IBD sharing around 0.5 for DZ twins, and consequently much less contrast/information for estimating V_A.
Note that the addition of measured genotype data like this to classical twin data does allow you to estimate more parameters than just A, C and E, and/or do some cool things with the data. If you are interested, see an example here:
Dave