flip a subset of the matrix before clustering rows about enrichedheatmap HOT 5 CLOSED

I finally have time to look at this comment. I quickly went through the method of this CAGT tool. For me, the algorithm basically contains two steps:

1. PAM partitioning on the normalized matrix with a large initial number of groups k
2. merging groups interatively with or without flip the matrix by the anchor targets.

For the first point, I think choosing optimized number of groups is a common problem for clustering and there is no perfect solution for this. And for the second point, I think it is worth to implement it in EnrichedHeatmap because it is true that only the relative distance to targets is important while whether it is on the upstream or downstream is of no importance.

from enrichedheatmap.

I've added a flip_upstream argument to normalizeToMatrix(). The process is basically:

1. normalize to the matrix with distinguishing upstream and downstream
2. flip the upstream part and add to the downstream

from enrichedheatmap.

Thanks Tommy! The closeness distance proposed in my paper is kind of a try to capture the patterns of the closeness of regions. It works, however I would say it is not an efficient way for doing it. I will read the paper you attached!

from enrichedheatmap.

the code in the paper was written in matlab, I only know R...I want an R version of something similar. look forward to reading your paper!

from enrichedheatmap.

Thanks will take a look and close it for now!

from enrichedheatmap.