**Figure 5:** Exploring the e_ect of shrinkage parameter (prior.df) choice on the di_erential expression of EV (orange) when gene signi_cantly di_erentially expressed
only with prior.df = 1. Two di_erent shrinkage values (prior.df = 1 and prior.df = 10) are chosen, and compared using the change in p-values (5(a)) and zoomed view
(5(b)), BCV (5(c)) and zoomed view (5(d)). Yellow indicates gene under investigation, chosen by mouse action on the plot. The interaction plot (5(e)) for this gene, and
the model estimates are shown (5(f)). Bars in the model estimates plot (5(f)) show the changes in di_erent dispersion estimates, for tagwise estimates, number means
prior.df. The gene investigated here becomes less signi_cant from the _rst shrinkage value (1) to the second (10), as seen that it has a smaller p-value on the horizontal
axis (< 0:05) and higher on the vertical axis (> 0:12) (5(b)), and the increase in tagwise dispersion (5(f)) with prior.df = 10. It makes sense because the variation in
RPA (green) is larger than for EV (orange) (5(d) and 5(e)), and trended dispersion is bigger. EV should really be considered more signi_cantly expressed. Shrinkage
increases the inuence of RPA dispersion on the EV dispersion estimate. |