Re: [AMBER] Help: Clustering trajectories based on dihedral angle?

From: Daniel Roe <>
Date: Fri, 25 Mar 2011 21:43:59 -0400


On Fri, Mar 25, 2011 at 6:35 PM, Thomas Cheatham III <> wrote:
> dihedral angles) however better would be to look in the AmberTools manual
> for the command called "dihedralcluster". áThis has a different syntax and
> clusters dihedrals.

I actually made a mistake in the manual, the actual command is
"clusterdihedral" (to avoid parser confusion with the 'dihedral'
command). However, I'm not sure it made it into the AmberTools 1.4
manual, so here is the full description of the command:

clusterdihedral out <filename> cut <clustersize_cutoff> framefile <filename>
                     clusterinfo <filename>
                     [dihedralfile <filename>] | [phibins <bins>
psibins <bins> <mask>]

Cluster frames in a trajectory using dihedral angles. To define which
dihedral angles will be used for clustering either an atom mask or an
input file specified by the dihedralfile keyword should be used. If
dihedralfile is used, each line in the file should contain a dihedral
to be binned with format:


where the ATOM arguments are the atom numbers defining the dihedral
and #BINS is the number of bins to be used (so if #BINS=10 the width
of each bin will be 36║. If an atom mask is specified, only protein
backbone dihedrals (Phi and Psi defined using atom names C-N-CA-C and
N-CA-C-N) within the mask will be used, with the bin sizes specified
by the phibins and psibins keywords (default for each is 10 bins).

Output will either be written to the terminal or the file specified by
the out keyword. First information about which dihedrals were
clustered will be printed. Then the number of clusters will be
printed, followed by detailed information of each cluster. The
clusters are sorted from most populated to least populated. Each
cluster line has format

Cluster CLUSTERNUM CLUSTERPOP [ dihedral1bin, dihedral2bin ... dihedralNbin ]

followed by a list of frame numbers that belong to that cluster.

If specified by the framefile keyword a file containing cluster
information for each frame will be written with format


where DIHEDRALBINID is a number that identifies the unique combination
of dihedral bins this cluster belongs to (specifically it is a
3*number-of-dihedral-characters long number composed of the individual
dihedral bins).

If specified by the clusterinfo keyword a file containing information
on each dihedral and each cluster will be printed. This file can be
read by SANDER for use with REMD with a structure reservoir
(-rremd=3). The file, which is essentially a simplified version of the
main output file, has the following format:


dihedral1_atom1 dihedral1_atom2 dihedral1_atom3 dihedral1_atom4





If a cutoff is specified by CUT only clusters with population greater
than CUT will be printed.

Hope this helps.


>> 2) áIf clustering is based on the rms deviation can we use reference
>> structure for clustering. For example, i want to make 3 clusters and i have
>> 3 reference structures. Based on this reference structures' rmsd can we
>> cluster the trajectory?
> The final section of the "cluster" command in the manual mentions the
> "decoy" option with the means clustering algorithm.
> cluster out decoy-test representative pdb average pdb means clusters 3 \
> árms decoy decoys.trj .CA
> To do this, create a AMBER trajectory file from the three reference
> structures (decoys)
> trajin decoy1.pdb
> trajin decoy2.pdb
> trajin decoy3.pdb
> trajout decoys.trj nobox
> --tec3
> _______________________________________________
> AMBER mailing list

AMBER mailing list
Received on Fri Mar 25 2011 - 19:00:02 PDT
Custom Search