Comparison of the global gene-expression profiles of normal and 5q− CD34⁺CD38⁻Thy-1⁺ and CD34⁺CD38⁺Thy-1⁻ cell populations. (A) Number of genes versus minimal Fisher score for 5q− CD34⁺CD38⁻Thy-1⁺ candidate HSCs when compared with normal CD34⁺CD38⁻Thy-1⁺ HSCs (solid line), normal CD34⁺CD38⁺Thy-1⁻ progenitors (dashed line), and 5q− CD34⁺CD38⁺Thy-1⁻ progenitors (dotted line). (B) Number of differentially expressed genes with Fisher score above 2 for 5q− CD34⁺CD38⁻Thy-1⁺ candidate HSCs when compared with normal CD34⁺CD38⁻Thy-1⁺ HSCs, normal CD34⁺CD38⁺Thy-1⁻ progenitors, and 5q− CD34⁺CD38⁺Thy-1⁻ progenitors, respectively. Error bars indicate 95% confidence interval (Poisson statistics). (C) False discovery rate as a function of number of top-ranked genes using the Fisher score. Results are shown for 5q− CD34⁺CD38⁻Thy-1⁺ candidate HSCs when compared with normal CD34⁺CD38⁻Thy-1⁺ HSCs (solid line), normal CD34⁺CD38⁺Thy-1⁻ progenitors (dashed line), and 5q− CD34⁺CD38⁺Thy-1⁻ progenitors (dotted line). The false discovery rate (ie, number of accepted genes in a permutation test divided by the same number in the correct list) was high when comparing 5q− HSCs with normal HSCs but essentially zero when compared with normal progenitors. A high false discovery rate implies that the compared populations do not differ in many more genes than expected by random fluctuations, whereas a low false discovery rate implies significant differences in gene expression between the compared cell types, reflecting more extensive differences than expected by random fluctuation.