Cluster Trees, Bx-Python: Is It Possible That The Number Of Clusters Increase As Cluster_Distance Increases?

Question

[Click downvote]

I have found the following behavior in bxpython which sounds like a bug to me:

I have a loop where I create a new Interval tree and check how many clusters it contains for the same data. It outputs:

With max_gap: 0 #clusters is: 156

With max_gap: 1000 #clusters is: 151

With max_gap: 2000 #clusters is: 155

With max_gap: 3000 #clusters is: 155

With max_gap: 4000 #clusters is: 156

With max_gap: 5000 #clusters is: 158

Is this possible? I thought that number of clusters should decrease monotonically with the length of cluster_distance?

Just wanted to double check before I report this as a possible bug.

The relevant code is:

max_distance = [i for i in range(0,5001,1000)]
for max_gap in max_distance:
    temp_tree = build_cluster_tree('chrX2.map' 10, max_gap)
    print "With max_gap:", max_gap, " #clusters is:",len(temp_tree.getregions())

More infor about Bx-pythons clustertrees here: Finding and displaying short reads clustered in the genome

[Ben]

As an aside: list comprehensions are cool and all, but range() outputs a list anyway.

Answer 1

[Damian Kao]

It might have something to do with the minimum number of reads you defined to be considered a cluster. As you increase the cluster distance, you might be adding more reads to previously discarded clusters, allowing them to pass the minimum reads check.

I wonder if you can relate this monotonic trend to distribution of reads across your reference. A set of reads that are very evenly distributed would probably show a large number of clusters initially and a sudden decrease to a single cluster after a certain cluster distance. Whereas something that is not evenly distributed will probably show a moderate amount of clusters initially and a slow decrease to single cluster as you increase cluster distance.