Clustering regions in plotHeatmap

[rikrdo89]

I have a clarification question. I have two BW files representing signal tracks of my control and treatment samples, and a bedfile containing my region of interest. Could I use the clustering methods in plotHeatmap to classify my regions into 3 categories: 1. those that gain signal, 2. those that lose signal, and 3. those with no change?
I was a bit confused by the docs (https://deeptools.readthedocs.io/en/develop/content/tools/plotHeatmap.html) which states that the --kmeans and the --hclust parameters Only works for data that is not grouped, otherwise only the first group will be clustered. . I am not sure what is meant by "grouped data".

Can someone suggest what would be the ideal parameters for clustering regions in the way I am describing? Does the --silhouette parameter is helpful in this case?

Thanks a lot!

I included a representative image showing that, no matter how many clusters I set using kmeans or hclust, I cannot uncover regions with differential signal enrichment (see arrow). It seems like the algorithm is trying to cluster based on similarities across samples, and gross differences in signal profile, but it is not helping me identify regions that are differentially enriched.

[miaoxiuxiu]

Using R package DiffBind to identify regions that are differentially enriched.
And plot heatmap using deeptools.