I did several HADDOCK runs using different parameters and I’d like to see how similar are the models obtained in these different runs. For this I’m clustering all obtained models (basically, cluster*_1.pdb files) using the FCC clustering (GitHub - haddocking/fcc: Fraction of Common Contacts Clustering Algorithm for Protein Structures). The reported coverage is between 34% (for 0.75 cutoff) and 57% (for 0.33 cutoff).
My question is: Why are not all models clustered? Are some models so distant from the others so they can’t form a cluster of even 2 members?
Oh, I think I’ve found the answer - setting the minimum cluster size (for cluster-fcc.py) to 1 (option: -c 1) results in 100% coverage, with some clusters containing only 1 model (i.e. no neighbours).
But then effectively you are ranking based on the individual model scores and no longer on the cluster-average score (over the best 4 model of the cluster if minimum four members are required)
Yes, true, but I tried to cluster only the cluster*_1.pdb files which had significantly different subunit-subunit orientation (the cluster*_1.pdb through cluster*_4.pdb files from the same cluster looked very similar).