Decorrelations do not come from the same samples. Plots of (a) sulfate transporter (SLC26A2) vs. lysine biosythesis pathway; (b) fatty acid binding protein (FABP5) vs. lysine biosythesis pathway; (c) fatty acid binding protein (FABP5) vs. sulfate transporter (SLC26A2). Loess curves and rug plots are given as in Fig. 4. Comparison of the plots reveals that the tumor cell samples responsible for the loss of correlation in sulfate transporter vs. lysine biosynthesis are not those which also exhibit loss of correlation in fatty acid binding protein vs. lysine biosythesis; it is clear from (c) that amongst tumor samples, high levels of expression in one gene do not necessarily correspond to that in the other, suggesting that the outlying points in plots (a, b) do not originate with the same sample.