Report a bug about cluster distance

[zhangfei1947]

Dear developer,
Sorry to disturb, I found a bug in using starcode and it really confused me.
I ran this command "starcode -d 1 -t 4 --input b1.fa --print-clusters > b1.cluster"
but output is like this, it seems that distance configure didn't work?
CATACCAA 2310939 AAAACCAA,AAAACCAC,AAAACCAG,AAAACCAT,AAAACCCA,AAAACCCG,......
...
...

[ezorita]

Hi, thanks for reporting.

Well, this is not a bug, this is how starcode is supposed to work. What starcode reports at the output is the result of clustering the sequences after matching them at a specified distance. So, to make things simple, there are two steps:

1. In the first step, starcode finds all the pairs of sequences that match each other at distance 1.
2. At the second step, it creates a network with the matching sequences and clusters the sequences following the specified algorithm (message passing by default).

That means that even if you set -d to 1, some clusters may contain sequences that are more than 1 mismatch from each other, especially when the input data set is too dense (almost all the combination of nucleotides are present).

To prevent this you can set a more restrictive clustering ratio, but that really depends on the nature of the biological data you are trying to cluster.

I can try to help you further but I would need more information about the sequences you are feeding to starcode.

Here is an example of what is going on

Say that you have an input with three sequences:

1. ATTTGAC
2. ATTCGAC
3. ATTCCAC

We set starcode to find matches at distance 1, and finds the following matches:

1 matches 2 at distance 1
2 matches 3 at distance 1

So the network is:

ATTTGAC <->  ATTCGAC <-> ATTCCAC

Which results in a cluster where ATTCGAC is the centroid and contains the three sequences, even though ATTTGAC and ATTCCAC are at distance 2.