Ciao all,
I do not know if this is the most appropriate section for posting however I will give it my best shot. I am conducting a capstone project for my genetic epidemiology program but unfortunately my mentors are unaware of the methods used for the project I will describe here so they encouraged me to connect with individuals on this website. Some may argue I should do a project that I can receive some assistance with from my mentors but I am passionate about this project!
I have a sample of twins raised apart and twins raised together. It includes the outcome I examine which is a measure of depression (it is continuous but also can be categorized in 3 categories). The sample also includes measures on the guardians who raised the individuals (some are binary like whether there was smoking in the household and some are continuous like income in the household). The objective for this project are threefold- firstly, to estimate how heritable depression is and relate it to the literature (the sample has monozygotic and dizygotic twins raised apart and twins raised together). I believe this is viable through intraclass correlations but I do not know how to make efficient use of the sampling design. Next I want to test for relationships of each of the guardian measures and depression to see how they may have an impact. Thirdly I want to test for mediators like social support.
This is my first time conducting such a rich empirical study. It is perhaps ambitious but I have done much reading and now am needing assistance with how to implement these analyses into R. In my preliminary efforts and literature investigation I do believe I want to use a SEM or fixed effects framework in favour of the ACE approaches. Feedback is so appreciated.
Cheers,
Oscar
Printer-friendly version
Hi, Oscar. I'm curious how large your sample is, and what proportion of its twin pairs are reared-apart? Also, what country or countries are your twins from?
It is certainly possible to calculate a heritability estimate from the intraclass correlations calculated from twins. Indeed, the intraclass correlation for MZ twins reared apart directly estimates the broad-sense heritability of a trait. But, ANOVA estimation of heritability is a bit old-fashioned. Consider instead using maximum likelihood (such as with OpenMx).
Keep in mind that characteristics of the rearing guardians may appear more strongly related to the child's phenotype when the child is the biological offspring of the guardians, due to passive gene-environment correlation.
Again, how large is your sample? Small samples are often underpowered for mediation analysis.
So, how much progress have you made so far?
Could you possibly point me in the direction of doing a simple univariate analysis using maximum likelihood in OpenMX to estimate hertiability and also the correlation within twin pairs such as for education all with confidence intervals?
Try adapting the attached script for your purposes. The script is written for one quantitative phenotype. One major change you'll need to make is that this script is written for twins reared together, but you'll have four groups instead of two: MZA, DZA, MZT, and DZT. Also, you'll be able to simultaneously estimate variance due to additive genetics (A), dominance (D), shared environment (C), and nonshared environment (E), if you want. This script currently adjusts for the effect of age on the phenotype; you will probably want to also adjust for sex as well.
The models in this script also impose an intraclass structure by constraining the model-expected phenotypic variance (and the regression equation for the phenotypic mean) equal for twin #1 and twin #2, in both zygosity groups. You may first want to fit less structured models to your dataset, perhaps to calculate descriptive statistics or check for anomalies in your data. Such models are unlikely to be of substantive interest to you, though.