Anyone knows how to use OpenMx for twins study including non twins siblings, like family model?
Thanks
Main menu
You are here
how to model Twins and non twins siblings
Tue, 01/05/2016 - 00:41
#1
how to model Twins and non twins siblings
Printer-friendly version
The simple answer is that you can treat non-twin full biological siblings as if they were DZ twins if you want. Alternately, you could use non-twin full sibs to identify an additional biometric variance component: so-called "twin effects," which contribute to the MZ and DZ covariance, but not to the non-twin sibling covariance. In any event, remember to statistically adjust for age and sex in your analysis.
With additional details from you, or with a more specific question, we could give a better answer.
Thanks a lot, The "twin effects" alternative seem very interesting, do you know how to model it, or is there any sample script for that?
For detail: I 'm analysing twins brain scans database (from the Human Connectome Project), it consist of ~500 subjects, about half twins (DZ/MZ) and half their non twins siblings. I try to estimate heritability on a sets of metric derived from brain functional connectivity.
Offhand, I don't know of any sample script for it. But it would be pretty straightforward to adapt a script that does the ACE or ADE model with twins (e.g., this). You would have three kinship groups: MZ twins, DZ twins, and full sibs. There would be an additional variance component that contributes to the MZ and DZ twin covariance, but not to the full-sib covariance.
There are some materials here: http://www.vipbg.vcu.edu/academics/courses/course-materials/hgen619-2015/ - a powerpoint and some scripts. Might be some more at the Boulder workshop site, e.g., https://ibg.colorado.edu/dokuwiki/doku.php?id=workshop:2014:schedule or in other even-numbered workshop years at that site.
So you want MZ and DZ groups that contain more than two people, with the first two being the twins, and the rest (up to max number of non-twin siblings in a family) are non-twin sibs).
It's not super common in data-collection: often just a handful of sibs, hence script libraries tend to concentrate on 2-member families.
Of course these extra sibs are very valuable (3 or even 6 or more covariances per family instead of just 1), but you need to expand the matrices holding these accordingly.
The DZ group covariance algebra will look like three (or more) people each sharing half their genes).
The MZ group covariance algebra will look like person 1 and 2 sharing 100%, and then the rest 50%.
Important as RobK says to include age and sex as covariates (preferably in the script, but in often implemented as a regression prior to analysis (make sure the regression is on the sample as a whole, not group by group)) as you now have different ages in a family, which will lower covariance artificially (of course the identical age (and sometimes sex) of the twins raises their correlation with one another also, which is why age and sex are always controlled).