This rule is dubious: added as a quick fix for expected inference in GO-CAM. The problem is most acute for transmembrane proteins, such as receptors or cell adhesion molecules, which have some subfunctions inside the cell (e.g. kinase activity) and some subfunctions outside (e.g. ligand binding). Correct annotation of where these functions occurs leads to incorrect inference about the location of the whole protein. This should probably be weakened to "... -> overlaps"