Figure 1

The effect of S₀. With real microarray data the absolute value of the t-statistic often shows erratic behavior for small values of the standard error S, with an increased risk of false positives. By choosing the constant S₀ in equation (1) wisely one can alleviate this problem. In the right panel we see that the statistic d in equation (1) downplays the importance of some of the genes with low standard error, compared to the t-statistic (left panel). Data from Golub [16] were used, and S₀ was chosen as the 5% percentile of S values (see also Discussion).