Segmentation Tutorial Part 7 - veeninglab/BactMAP GitHub Wiki

Statistical tests

To check whether the different conditions are similarly distributed, I chose to do a Kruskal Wallis test followed by a Pairwise Wilcoxon Rank Sum Test, which is, when adjusted, suitable for multiple testing. Here, I only test the diffence in cell length, but you can do this for cell width and area too.

onePerCell$condition <- as.factor(onePerCell$condition)

kruskal.test(max_um ~ condition, data = onePerCell)

##
##  Kruskal-Wallis rank sum test
##
## data:  max_um by condition
## Kruskal-Wallis chi-squared = 888.3, df = 4, p-value < 2.2e-16

pairwise.wilcox.test(onePerCell$max_um, onePerCell$condition, p.adjust.method="BH")

##
##  Pairwise comparisons using Wilcoxon rank sum test
##
## data:  onePerCell$max_um and onePerCell$condition
##
##                      Clement_Oufti Jun_Oufti Jun_SuperSegger
## Jun_Oufti            <2e-16        -         -
## Jun_SuperSegger      <2e-16        0.3370    -
## Lance_MicrobeJ       <2e-16        <2e-16    <2e-16
## Renske_Morphometrics <2e-16        <2e-16    <2e-16
##                      Lance_MicrobeJ
## Jun_Oufti            -
## Jun_SuperSegger      -
## Lance_MicrobeJ       -
## Renske_Morphometrics 0.0039
##
## P value adjustment method: BH

There can be many reasons why the segmentation is so different per cell. The program might have a default cell width setting (as Oufti has, for instance), making the cells extra fat or extra small. One program might split the cells earlier than others. But people can also influence the outcome (In the case of cell length, the two more similar results are both from Jun!). You might correct segmentations differently or put the threshold a bit higher than someone else. I think the take home message is to make sure that if you do cell size analysis, don’t assume you can compare your output one on one with someone else, even if you use the same program.

---

:arrow_left: Segmentation Tutorial part 6: Compare segmentations visually	Segmentation Tutorial part 8: Extra's :arrow_right: