I am attempting to conduct a genomic SEM with a GxE term and a term for the effect of G on E. The model would also include 11 other covariates. The study was a case-control design.
I used the R package gwsem to conduct a GxE analysis without the effect of G on E. However, I can not use it for the current analysis since the package does not permit a term to be a covariate and a dependent variable.
I have tried to create a custom model by writing mxPath statements but I ran into two problems. One, I am not sure how to specify the interaction term. Two, I get an error message when I run mxRun that the E term is not specified although it is included in two path statements: once as from=E and once to=E.
I would greatly appreciate any help in specifying my model.
Printer-friendly version
In general, you cannot use the same variable as a covariate and a response. I don't really follow what you are trying to accomplish. Maybe it would help us if you shared your complete code?
Let me try to clarify what I'm trying to do. The phenotype is the number of symptoms a participant experiences, so the outcome is quantitative. The environment (E) is whether a study participant has used cannabis before age 15, so this variable is binary. My colleagues and I hypothesize that SNP (G) can have a direct effect on the phenotype (number of symptoms). Past literature shows that E (cannabis use) can also have a direct effect on the phenotype. In addition, we want to test the interaction between G and E. So far this a typical GxE, and I can test this model using the gwsem package.
The added complexity is that we want to test whether G has a direct effect on E, for there is some evidence of a genetic predisposition for earlier cannabis use. It is this last element that I am having trouble building into the model.
I have attached the script I have so far as well as a diagram of what I am trying to model. Since I have not written script for the GxE interaction, I have drawn that path in the diagram with red.
Again, any help with this is greatly appreciated.
It sounds like you want to do both moderation (GxE) and mediation (indirect effect of SNP). Correct?
Yes, that is correct.
Seems that if you make
data$modSNP <- data$SNPyou could use data.modSNP for the moderation but still have the SNP in the model as an observed variable. Always beware of uneven measurement precision in your outcomes when testing moderators. Such heteroscedasticity can fool analyses into thinking that there's GxE when there isn't, or that there isn't when there is...You can estimate the SNP x Use effect with reference to the attached diagram. Data.use is the definition variable for use. Pointing an arrow to another arrow does not a proper path diagram make.
You might try using gwsem GxE. Will that get you the model you want?