This sounds like an interesting idea, but I cannot think of a good way to color code all possible GO terms in a way that is accessible to human perception. In a scatter-plot of a PCA you might as well end of with no visible structure, that will make it impossible to spot any groups, because most genes are not differentially expressed or have no role under the experimental factors; in the same way most GO terms should not be affected under any given experimental conditions. => only few genes AND only few GO terms are significantly altered at any time. But, if you are not afraid of whipping some visualization up for yourself (which potentially could lead nowhere), you could do the mapping once you focus on the enriched set of GO terms under the process you are studying. You could try the following:

• Get some gene sets, e.g. by DE analysis
• Run GO term enrichment analysis on the sets
• Further reduce the number of relevant terms using e.g. REVIGO tree maps
• assign a color to each gene, based on it being assigned to a child node of the few relevant terms
• chose the color based on semantic similarity (e.g. SimRel) such that similar terms have similar colors, e.g. use the same colors as in REVIGO treemaps

I am sorry, but I don't have an easy code recipe for the last 2 steps. Hope that helps.