I have a large distribution of paired percentages. Is there a best way, or any way, to test for sig. different pairs?

#1
These percentages are based on RNA sequencing count data for which replicate values have been merged and transformed to these percentages (which represent average lengths, not counts). But since the replicate data has been lost in the transformation to %, I'm not sure if I can still test for significant differences between the %age values of the conditions in a way similar to a standard program called DESeq that finds significant differences between the original count data between conditions.

Is it still possible to obtian a meaningful p-value for each pair of %age values based on the average difference in pair values we see across the entire set? As in, what are the significantly different outlier pairs? Sorry I'm in a bit over my head. Thanks for any advice.

Example data:
condition 1 condition 2
0.45 0.37
0.2345 0.346
0.3456 0.12
0.8756 0.9767
0.06 0.17
.
.
.
etc ~ 5000 pairs
 

katxt

Well-Known Member
#2
I suspect you might have a hard job ahead. Have you taken a quick look by graphing the differences? A histogram or boxplot or probability plot?