numberclust question

[kyrgyzstanygnacio]

Hello,

I have a question about the numberclust parameter. I am trying to use it filter out small singleton clusters, however it seems to be filtering out everything.

Here is one part of my visualization with numberclust set to the default value with no filtering. In the figure below, there is a cluster of 4 bins on the righthand side that look significantly different between conditions.

But when I set numberclust to 2, this group of bins and others like it are being filtered out and the resulting visualization shows no significant regions. I am not using the distclust parameter.

Am I misunderstanding how numberclust works? And if so, is there a way to easily remove these singleton bins without discarding larger clusters, as seems to be happening here?

Thank you!

[ay-lab]

So, in the background it uses --distclust to first cluster the differential regions and then measure the number of clusters. The numberclust is used as filtering criteria to exclude regions that are not satisfied. I would suggest you to set --distclust 0 , this will cluster differential compartments that are adjacent to each other and try --numberclust 2 and see if its working or not.

A simpler option we later figure out is to use bedtools merge option on the output of dcHiC. Using the bedtools merge -d -c -o option you can set any distance to cluster (-d) and cluster the columns (-c) based based on different kinds of operations (-o)