...

To understand how including a Gene*Factor interaction may improve the fit of the model, consider the complex GO ANOVA design in the case of a dose-time analysis of a drug. While it may seem clear that the ANOVA factors in the middle box - dose, time, and the dose*time interaction should be specified (to consider the effect of dose, time, and the change in the effect of dose over time) what to put in the rightmost Gene*Factor box is not as clear. Adding dose alone (which is actually Dose*Gene) will check if different drug doses affect the pattern of gene expression. Similarly adding time into the right box (which is actually Time*Gene) will identify gene ontology categories that are affected in different times but differentially across the genes. While this may be the true limit of questions of interest, including the interactions of the gene and both dose and time may be prudent. In general, if it is likely, or expected, that a factor will affect gene distribution within functional categories, then the factor should be included in the Disruption Factor(s) box if the gene distribution is being analyzed at all.

...