Kruskal Wallis test on z-scores

[PleaseHelp]

Hi all,

I would like to check if amounts of different proteins in my samples are changed over time. For this I would like to normalize the different measurements by z-score. Afterwards, I plan to include this z-scores in a Kruskal Wallis Test to detect significant changes.
Is this legitime?

Thanks for all help,
Steve

[hlsmith]

Clarification, so you have longitudinal (perhaps pre and post) data, where you used two different measurement instruments and want to test for change between times using Kruskal Wallis Test.

[PleaseHelp]

Hi hlsmith,

thanks for your answer!
Yes, I've longitudinal data. Protein identification and quantification was done with the same instrument using only 3 biological replicates (because it's expensive) of samples harvested at 5 different time points. I am now interested in proteins with significantly increased or decreased amounts over time. Z-score normalization works much better than median or total normalization, that's why I want to use it. Because obeserved amounts are not normal distributed, I would use Kruskal Wallis Test.
Is this compatible?

Thank you so much for your help!
Steve

[hlsmith]

You slightly confused me by providing a little context info, which usually does the opposite. So if I got this right these measurements come from the exact same subjects of three groups at two times and you used the exact same instrument?

The Kruskal is the non-parametric test for continuous data that is not normally distributed. So if your data is not normally distributed and you have three groups, your data seem to fit the broad requirements of the test. Read about the test to see if there are any assumption you need to consider. If you find significance in this omnibus test, then you would perform Wilcoxon-Mann-Whitney tests (non-parametric equivalent test to the t-test) to examine differences between the three groups, and would want to control for pairwise error (multiplicity).

Though if you use the same test twice, you may not need to Z-score normalize these data.

[PleaseHelp]

So, I try to explain it in more detail. I measured protein concentrations in samples of five different time points. For each time point three comparable samples (replicates) were measured. The measurement is slightly different on each run which I want to compensate by normalization. Of I understand you right, you recommend the kruskal wallis test to detect global changes in the protein amount and a post hoc analysis (eg withney test) to see when the significant change exactly happens. Then no normalization would be required. Do I understand you right?

Thank you for all your help!

[hlsmith]

So you have 5 time points for each member of your sample? That is sounding more like time-series then the two time points for the Kruskal-Wallis.