compute center for each treatment

[eschen42]

Many multivariate procedures such as PCA suggest through clustering of samples how treatments might differ, but they are not as helpful in showing similarity the way that a heatmap does. However, producing a heatmap based on individual samples intermingles the samples for different treatments, making it more difficult to surmise quickly which treatments are more similar in effect.

It may be useful to identify the "center" of the samples for each treatment and look at the relationship among the centers. There are several conceptual centers from which to choose. This issue is open to add the following options for centering:

• none - Do not compute centers
• centroid - For each treatment, compute the center as the vector comprising the mean for each feature.
• median - For each treatment, compute the center as the vector comprising the median for each feature.
• medoid - For each treatment, compute the center as the medoid, i.e., the sample that is most "central".
  • Because this calculation is sensitive to correlation among features, perform this computation in principal components space.
  • An elegant solution in R for identification of the medoid is presented here

[eschen42]

Addressed in release v0.98.15

[eschen42]

Closing because release has been published.

[eschen42]

Data should be scaled as part of PCA for more consistent distance measurement: Have prcomp standard-scale the data before performing PCA.

[eschen42]

bug addressed in release v0.98.16