Mathematically it is not possible to encode N+1 continuous dimensions into N continuous dimensions.
If the dimensions were discrete then you could use a variation of Cantor's diagonalization technique to map any 4D point to 1D (and therefore trivially to 2D). One way of thinking about this is that if dimensions were discrete then you could map each location to its matlab linear index.
So you can do it if you are willing to to discretize your space.
The next question to ask is whether the resulting graph would be understandable to humans. The answer to that is "Probably not." But that is a different question than whether some encoding technique exists.