Consensus heatmap?

[jarbet]

A cool feature to add to the package would be a consensus heatmap to visualize the consensus matrix, and you could add a covariate bar for cluster assignments.

For example, something similar to this (example from the cola R package):

• so the main heatmap would show the consensus probabilities for all pairs of subjects (high is blue, low is white)
• Then you can add a covariate bar for the cluster assignments (see the "Class" bar on the left, you can ignore the other bars)
• Ideally, you want to see diagonal dark blue boxes corresponding to the number of clusters, and the off-diagonals should be white

Then it would be really cool to be able to make the heatmap for a different number of cluster assignments (e.g. ranging the total number of clusters from 2 to 6), like here:

• Unlike cola, the consensus matrix for IMPACC will be the same for all K, but nevertheless it would still be cool to add the cluster-assignments for varying K to the heatmap. Then maybe such plots could help the user select the best value of K.
• You might need to sort the consensus matrix in a clever way for a given value of K (e.g. sort patients by cluster assignment, for example) in order to see the clustering patterns better

[luqingan]

Hi @jarbet ,

Thanks a lot for the great suggestion! I'll work on it.

Best,
Luqin