[gmx-users] clustering dynamical structures

[Mark Abraham]

Dmitri Dubov wrote:
> Dear GMX'ers,
> 
> Many gmx tools (g_rms, g_rmsdist, g_cluster etc.) calculate RMSD of distances. Take some molecule having two docking sites (A and B) and hydrated with two water molecules (C and D). Let's inspect two structures realized. The first structure is with C docking to A and D to B, and the 2nd one has C docking to B and D to A. These structures have large RMSD of distances between FIXED PAIRS OF  THEIR ATOMS, and to my knowledge g_cluster will differ them. How can I cluster these structures?

You should test your hypothesis with g_cluster by converting two such 
frames to a .pdb "trajectory" with trjconv, and then editing by hand to 
permute such a pair, and comparing the results before and after permutation.

> (Actually it is a very simplified example of my problem. My need is in clustering structures for system with many permutations.)

It's hard to advise when we don't know your real objective. Do you want 
to cluster over fully solvated structures, but losing fixed-pair mapping 
of waters? If so, then that's going to require quite expensive 
permutation over all possible orders of waters to find the global 
minimum RMSD. Heuristics exist to find a close-to-minimum permutation, 
but guaranteeing global minima is hard.

On the other hand, if you merely want to cluster with any waters in the 
docking sites permuted, and ignore the rest, then you may need to script 
g_dist using each water molecule as a group and a key atom in a docking 
site as a group in order to identify the water molecule(s) in the site. 
Then trjconv with the same index file and that info can write a 
trajectory with just the selected waters. Now you may merely need to 
permute over a handful of possibilities with g_rms, before finally 
writing the permuted selected waters + solute to a trajectory (with 
trjconv). This is now prepared for clustering with g_cluster - the 
waters are ordered such that fixed-pair mapping makes sense.

Mark