Improving secretion of recombinant proteins from yeast and mammalian cells: rational or empirical design?