DireX - Tutorial
Kinked Helix at Low-resolution
This tutorial shows a very basic example of the
usage of DireX to flexibly fit a protein structure into a low-resolution
density map. The fitting of this small helix is done in just a few seconds,
which allows to conveniently test the effect of the various input parameters.
A perfect α-helix (extended-helix.pdb
) built with Chimera is
used as the starting structure. This starting structure
serves in this case also as the DEN reference structure
(see Introduction to DEN Method
The target structure (target.pdb
) is a kinked helix that
has a root-mean-square deviation (RMSD) of about 4.7 Å.
We first create a 10 Å density map based on this (kinked) target
structure map with DireX.
The program call can be found in the run.sh
# Make density
direx -f mkdensity.par -pdb target.pdb -cur cur_mkdensity.pdb -curmap
kinked-density.mrc -map self -map_apix 2.0
This command writes the file kinked-density.mrc
The keyword "self
" to the -map
let DireX generate a density map from the input PDB file instead
of reading a density map file. The parameter -map_apix
the grid spacing of the resulting density map.
The other input parameters are described below.
The goal is now to fit the straight extended helix
into the kinked density map to
obtain a fitted structure that gets as close as possible to the target structure.
The actual structure refinement is then performed with (see
also the run.sh
# Run refinement
direx -f refine.par -pdb extended-helix.pdb -cur current.pdb
-curmap curmap.mrc -map kinked-density.mrc -ox s.xtc
The file refine.par
contains a list of input parameters.
A list of all existing parameters can be found in the file
) which is included in the DireX distribution.
The -pdb option reads the atomic coordinates from a PDB file and, in
addition also topology information, i.e. which atoms are connected
by chemical bonds. The bonds are detected using a distance
DireX determines the bond lengths and angles directly from
the PDB file and it does not optimize those values. It is therefore
recommended that you use a (energy minimized) structure with
correct bond lengths and angles.
Also, note that only lines starting with "ATOM" and "HETATM" are read, everything else
The file current.pdb holds the current coordinates and is updated after
each iteration. After the refinement is finished it therefore
contains the final fitted structure.
The final fitted structure has an RMSD (Ca-atoms) of 0.9 Å to the target structure.
The file mapcc.dat
contains the correlation between the model map and the
target density map. This file can be viewed e.g. with the plot program
xmgr (or grace).
The file traj.xtc
holds a trajectory of the atomic coordinates,
each frame of the trajectory
corresponds to one step during the iterative fitting. The file is in XTC format
which can be read e.g. by VMD:
vmd -f extended-helix.pdb traj.xtc -m kinked-density.mrc
The -m option reads the density map as well. This way you can watch the
structure as it is being fit to the density. Note that VMD does not update
secondary structure assignment by default (you can find a VMD script on the
VMD website that can do it, though).
An important part of finding the optimal parameters is to balance the
fitting to the density map and at the same time obtain a reasonable structure
in terms of local geometry, ramachandran statistics, and secondary structure.
The DEN restraints are the key element in DireX to control this balance.
In the next few steps, we will play with some parameters that define
the DEN restraints:
The number of restraints should be at least equal to the number of atoms,
but typically a better choice would be factor of two. In this case we
therefore use 412.
This is one of the key parameters in the DEN approach. In the present case,
the DEN reference model is the same as the starting model (extended-helix.pdb
A small γ-value will keep the structure close to this reference model, and
a large γ-value will allow larger deformations of the network and therefore
should result in higher map correlations. There is usually an optimum choice
that best balances reference information and density fit. In principle,
this optimum value should be determined by cross-validation (this is not yet
part of the tutorials).
Since we know the correct target structure for this case, we will compute
the RMSD (root mean square deviation) to the target structure to assess the
effect of the γ-value. The three plots on the right show the RMSD, the
fraction of residues that are within the allowed region of the Ramchandran plot,
and the density map correlation. From the RMSD plot, the optimal γ-value
is found to be 0.9.
The strength should always be high enough to keep the local geometry in shape.
A value between 0.2 and 0.4 is usually a good choice.
If you feel the structure does not move enough, even for larger gamma values,
you can decrease this value. However, if you have secondary structure information,
it might be better to use the den_loop
parameter, to decrease only the
strength of restraints within loops, this is described in more detail in the
Using a value larger than 0.5 is not recommended: the computation could become unstable.
den_upper / den_lower:
The DEN restraints are chosen randomly between atoms pairs that are
within the distance interval defined by these two parameters.
3 - 15 Å is the default, which should be fine in most cases.
The choice of the DEN restraints can be further restricted to atom
pairs that are within a given range of residue numbers, i.e. selected
atom pairs will