some comments

[MPietzke]

Nice app, well done! I may take some inspiration for my app, I like especially the remove outlier feature!
I found some issues / have some comments, maybe you appreciate some feedback (based on the online-app at: https://webapps.nutrimetabolomics.com/POMAShiny)

1. Upload data: “Exploratory report” button not working. Search button only searches for cell content, not column names, so it’s a bit useless.
2. Impute values: I’m not sure if the missing value imputation (KNN) is correct, does it include groups? For example for aspartate (in the 2 sample example) the highest value in the control group is 1.33, there are 3 missings that are filled with: 2.294, 4.395, 2.445, way higher than in the other samples.
3. Normalisation: This step requires the missing value imputation, correct? At least it’s not working when missing value imputation wasn’t performed. Here a warning would be nice.
4. Volcano Plot: Instead of a warning for 3 (or more groups) it would be nice to select just 2 of the groups for comparisons.
   p-value threshold is used as original p-value, before log transformation, while the FC input reads the logged value, which is a bit inconsistent. Thresholds shouldn’t be allowed to be negative. There should be a warning when points fall outside the x-axis range limit.
5. Density Plot: no x-axis labels, I guess it’s the normalized intensity?!
6. Heatmap: The group colours are random, which is a bit confusing, as you defined the colours for each group in the plots before. Moreover, when you change the settings (switch the names on or off) the colours change again
7. Cluster Analysis: The show clusters button is a bit useless? Here the clusters also swap when changing some parameters, due to the random nature of the kmeans. Better set the seed or try hierarchical K-means, which should give more consistent results.

I attach some screenshots:
Pomashiny issues.pdf

Further I am missing some more explanation, not every user might instantly know which missing value imputation or normalisation is the best for the dataset...

Best
Matthias