Rank Sum Tests
Hey guys,
im struggeling with some statistics given by the vcf file: the Ranksumtests. I started googleing arround, but that turned out to be not helpfult for understanding it (in may case). I really have no idea how to interprete the vcfstatisticvalues comming from ranksumtest. I have no clue whether a negative, positive or value near zero is good/bad. Therefore i
m asking for some help here. Maybe someone knows a good tutorialpage or can give me a hint to better understand the values of MQRankSum, ReadPosRankSum and BaseQRankSum. I have the same problem with the FisherStrand statistics. Many, many thanks in advance.
Best Answer

pdexheimer Member
The rank sum metrics use the Wilcoxon Rank Sum test, which tests whether or not two populations are the same (from the same distribution). The actual metric is the Zscore of the test, not the direct U score  see the "Normal approximation" section on the wiki page. The major takeaway is that values close to zero are "good" (less difference between reads containing the reference and alternate reads), while higher magnitude values reflect larger differences
Answers
The rank sum metrics use the Wilcoxon Rank Sum test, which tests whether or not two populations are the same (from the same distribution). The actual metric is the Zscore of the test, not the direct U score  see the "Normal approximation" section on the wiki page. The major takeaway is that values close to zero are "good" (less difference between reads containing the reference and alternate reads), while higher magnitude values reflect larger differences