Direction of Causation (DOC) models: When to use them?

According to Brad Verhulst, it's not worth using this model in psychiatric genetics where there are equate modes of transmissions

If one trait has a lot of heritability and the other has much less, then this model is a good candidate.

Andrew Heath et al. (1993) provides concrete guidelines.

One method of testing causality is Mendelian Randomization (see David Evan et al.s study).

Heath, A.C., Kessler, R.C., Neale, M.C., Eaves, L.J. & Kendler, K.S. (1993) Testing hypotheses about direction of causation using cross-sectional family data. Behavior Genetics 23: 29-50. [Fulltext PDF]

David Evan's paper
http://www.plosgenetics.org/article/info%3Adoi%2F10.1371%2Fjournal.pgen.1003919

Does epigenetics create significant bias in the ACE model?

The ITW 2014 workshop faculty agreed that epigenetics does not create significant bias in the ACE model. If epigenetics plays a role, it will show up systematically in all three components (A, C, and E) and there is no reason to belief that it would influence these components differentially.

Sometimes the standard error shows up as NAN. Why?

Sometimes the OpenMx output shows the standard error as NAN (not available). This happens usually when the model fails to compile properly. The algorithm cannot compute the standard error when it attempts to take the square root of a negative value.

One way to address the problem is to set a different starting value of the model and try again.

Good candidates for covariates

Gender, age, and genotypes are usually good candidates for coviariates. These variables are usually not the target phenotypes being measured. Modeling other phenotypes as covariates can be problematic.

Epistasis: A limitation for twin studies?

Several faculty members at ITW2014 mentioned epistasis as a potential limitation of class twin studies. Epistasis describes the potential possibility that the effect of one gene depends on the effect of one or more other genes. The question is, how problematic is epistasis for twin researchers?

Professor David Evans explained that epistatsis is usually now a serious problem (whewww!) because it happens very rarely and therefore it's highly unlikely that the particular phenotype you are studying suffers from epistasis.

#1 cause for OpenMx model results that are obviously wrong

Sarah pedland at ITW2014 shared this piece of wisdom - If your model results seem massively wrong, the first thing you should check is how your dataset is arranged. If you have variables in rows, and participate cases in columns, then that's the culprit. You must arrange your variables in columns and cases in rows. This dataset error is the cause more than 90% of the time!

What can I use OpenMX/R for?

I've always known that OpenMX is the go-to tool for behavioral genetic analysis. What I didn't realize until ITW 2014 is that it's an extremely versatile tool that can be used for all sorts of analytical tasks! Here are a few examples:

1. Structural Equation Modeling
2. Genome-Wide Association Study (GWAS)

R can be used for:
Mendelian Randomization (which requires just one line of R code! using the Structural Equation Modeling (SEM) library, and instrumental variables analysis)

What other problems have you used OpenMx to solve?