Figure 7.

Assessment of biases of the correlation coefficients estimated from 19 groups of expression data.$B\left({\widehat{\rho }}_{\mathit{\text{xy}}}\right)=\sqrt{\frac{\sum _{i=1}^{19}{\lambda }_i{({\widehat{\rho }}_{\mathit{\text{xy}}}-{\widehat{\rho }}_{\mathit{\text{xy}},i})}^2}{19}}$ for (a)${\widehat{\rho }}_{\mathit{\text{xy}}}=r_{\mathit{\text{xy}}}$, the Pearson correlation coefficients estimated directly from the large expression matrix; (b)${\widehat{\rho }}_{\mathit{\text{xy}}}={\overline{r}}_{\mathit{\text{xy}}}$, the average of correlations between expression profiles within i = 1,19 groups;${\widehat{\rho }}_{\mathit{\text{xy}},i}$ is the correlation between expression profiles within each group.