I am using MetaboAnalyst to analyze data from a randomized crossover trial. Which (if any) of the data reduction approaches in MetaboAnalyst (e.g., PCA, PLS-DA, etc.) can take the repeated measures (i.e., data dependency) into account? If not possible to do in the web-based version of MetaboAnalyst, can the R code be customized to take repeated measures into account?

Thanks in advance!

Daria

Code the repeats as a design factor and use the **Statistical Analysis [metadata table]** for analysis. Keep in mind that this is an exploratory analysis, you can do the following analyses and visualization

- Univariate stats: Two-way ANOVA, linear modeling (there are some statistical consideration “blocking factor”)
- Unsupervised clustering: PCA, heatmap (visually examine the data dependency here)
- Supervised analysis: Random Forest (require large sample size)

Thank you very much for your reply, Jeff.

Regarding PCA, when you suggest to “visually examine the data dependency,” do you mean I should determine how close each pair of points are to each other if all participants underwent two conditions (i.e., a randomized crossover design)? And that if participants tend to cluster together, this is evidence that the data dependency is skewing the results?

We have a fairly robust design where our two conditions tend to cluster separately, but I have been unable to determine if PCA in general is an appropriate data reduction approach when a participant is represented in the data more than once.

Thanks again!