FindAllMarkers vs presto displaying a difference in top 3 markers and logFC values

[parmejohn]

Hi all,

I know from #4105 that speed is the main difference from running presto to the default method, but after reading their paper and all the previous discussions, I thought that there would not be a difference. Although, when I run FindAllMarkers without presto installed and compared it RunPrestoAll from SeuratWrappers (with the matching parameter values), the p-val and log2FC values were slightly different, and my top resulting 3 markers list differed by 1 out of 87 as well.

Is the reason for this down to implementation of the Wilcox test? Is one of these methods considered to be more robust/accurate?