An approach coloring the cubies.
Let's color the individual cubes red and blue. We paint them in a "checker" (the 3D version, that is) pattern, so that the corners are red and the middles of each sides are blue, the centers of each face red, and the center of the cube blue, like this (created with POV-Ray):
Now let's say we have a path that starts at a red piece (at a corner), and ends in the center, a blue piece. Let's count the number of red and blue pieces we need to go through. There are $14$ red pieces and $13$ blue pieces. Since we start at a red, and end at a blue, and we need a "Red, Blue, Red, Blue, ..., Red, Blue" pattern (since each color only has neigbours of another color), this is never going to work: since, for such pattern, and $14$ "Red"s we need $14$ "Blue"s, but we only have $13$! So this is not possible.
A (possibly equivalent) approach using graph theory.
Since you mentioned graph theory, let's look at it that way, too. We make a graph with $27$ points (one for each cube) and two points are connected if and only if the cubes they represent share a face. Note that this graph is bipartite, since it only contains even cycles. An image of the graph we get (created with GeoGebra):
Now note that on the left side (that's the side where the corners of the original cubes are) we have $14$ points, while on the other side we have only $13$ points. Now we need to find a path that starts on the left side and ends on the right side. But we have to go back an forth between the left and right side, so when we've reached all $13$ points on the right side, we still have one left on the left side. Thus, such path does not exist.