PaGE 5.0 Documentation - the t-statistic

The PaGE algorithm relies on some statistic to measure differential expression. In theory any statistic can be used and the confidence of the results will still accurate. However an unwise choise of statistic may have no power for detecting differential expression.

PaGE 5.0 offers two statistics, the t-statistic and the ratio of means of the two groups.

The statistic is chosen when you choose menu item Run PaGE.

Direct Design and Paired vs. Unpaired In the unpaired case the 2-sample t-statistic is used. In the paired case and in the direct comparison design case the 1-sample t-statistic is used. For more details, including precise definitions of these statistics, see the technical manual.

The Tuning Parameter: Different statistics are sensitive to different kinds of false positives and false negatives. The t-statistic is very sensitive to the small variances that occur when there are relatively few replicates. The degree of this sensitivity can be controlled with the tuning parameter, which is added to the denominator of the t-statistic. The more replicates there are, the less of an impact this parameter has. For few replicatres (5 or less), generally the power of the results go up as this parameter goes up from zero, and then back down as the parameter becomes large. Maximum power occurs with some moderate value of this parameter.

PaGE sets this parameter for you automatically at a moderate value, which is reported in the PaGE output. You can override this default value using the t-statistic tuning parameter option in the Configurations dialog box. For more on this parameter, see the technical manual.