|
Manual EM Docking Tutorial
|
| The tutorial shows the manual
placement of structures into EM maps "by eye" and their subsequent
refinement with the colacor tool. The
results
can be compared to solutions distributed with the tutorial software.
More documentation is available in the user
guide, in the methodology page, and in
the published articles. |
|
Content:
|
| Download
and Installation
Follow first the
simple registration
and download steps .
In addition to
the executables, the
Situs_2.5_manual_tutorial/bin
directory contains two data files:
- 0_map.situs:
A low-resolution map of ncd, a kinesin family protein.
- 0_docked.pdb:
Atomic coordinates of ncd roughly docked to the map.
In
the
following, we will perform various modeling tasks.
The user can compare all generated files to the files in the
"solutions"
directory.
|
Data
Flow and Design
The series
of steps and the utilities that are required for the docking "by eye"
are shown
schematically
in the following figure. Detailed program explanations are
given
in the user guide.
Schematic
diagram of colacor related
routines. The main data
flow is indicated by brown arrows. The visualization
(orange) for the rendering of the bead models requires a molecular
graphics viewer (we use here the VMD
graphics program, Chimera and Sculptor also support Situs
format).
Standard EM
formats are supported
and are converted to cubic lattices in Situs format. This is done with
the map2map utility. Subsequently,
the data is inspected and, if necessary, prepared for the fitting using
a variety of visualization and analysis tools. All Situs tools require one
volume and one PDB structure for the fitting. Atomic
coordinates in PDB format can be transformed to low-resolution maps, if
necessary, and vice versa, to allow docking of maps to maps or
structures to structures. The
resulting docked complex can be inspected in the graphics program.
|
| Manual Docking
In many fitting applications an expert
user may have a pretty good idea where to place a biomolecule.
Therefore, it is quite a popular approach to manually dock atomic
structures
into low resolution maps. In molecular graphics program such as VMD one can shift and
rotate a
loaded molecule and save the new coordinates.
The following
sequence of commands
in the VMD text console (cf. VMD user
guide ) will load the docked structure 0_docked.pdb and render
it in cartoon representation, coded by color. The script then instructs
VMD to render the file 0_map.situs:
mol load
pdb 0_docked.pdb
mol load
situs 0_map.situs
mol top 0
rotate stop
display
resetview
display
projection orthographic
mol modstyle
0 0 Cartoon 2.1 11
5
mol modstyle
0 1 Isosurface 20.0 0 0 1 1 1
mol modcolor
0 0 Structure
mol
modcolor 0 1 ColorID 8
|
Then select the menu Mouse ->
Move -> Molecule. You can now shift the PDB structure with the mouse
and rotate it by pressing the Shift
key; the new coordinates can then be saved to a file by selecting the
VMD menu File -> Save
Coordinates (select "all" atoms). Play around with this and create two
new PDB files, 1_shift.pdb and 1_rot.pdb. The structure should be
halfway outside the map in the former, and rotated relative to the map
in the latter.
The results
of the 3 cases (when rendered with VMD scripts equivalent to the one
above) for
0_docked.pdb, 1_shift.pdb, and 1_rot.pdb should look like the following
figure:
(Click image to
enlarge)
You can be more aggressive with
displacing the structure, but then it is less likely that colacor will find its way
back to the correct local maximum of the correlation.
|
Refinement
We support this
manual docking by providing a tool, colacor, that calculates the
cross correlation (as a way to provide quantitative feedback) and
performs a single optimization run to the nearest maximum of the
cross-correlation coefficient (if a refinement after manual docking is
desired). As with colores above, the user
needs to select volumetric correlation explicitely (option
"-corr 0"), otherwise the default option (Laplacian filter) is
applied. colacor is essentially a
stripped down
version of colores, but it does not
center the input map
and PDB as in a global 6D search, instead it proceeds based on the
local geometry.
The first case above, 0_docked.pdb is
only slightly misplaced relative to the volumetric map. This is a good
start situation for manual
refinement with colacor. Here we give the results of a colacor
run based on these two input files (note the -corr 0 option). We assume a (default) resolution
15A in this case.
./colacor
0_map.situs 0_docked.pdb -corr 0
_____________________________________________________________________________
colacor> Options read:
colacor> Target resolution 15.000
colacor> Resolution anisotropy 1.000
colacor> Powell correlation algorithm determined automatically
colacor> Low-resolution map cutoff 0.000
colacor> Powell maximization ON
colacor> Grid size expansion factor 0.000 (thickness of additional
zero layer as fraction of map dimensions)
colacor> Standard cross correlation
colacor> Powell tolerance 1.00E-06 Max iterations 25
colacor> Powell trans & rot initial step sizes set to default
values
_____________________________________________________________________________
colacor> Processing low-resolution map.
lib_vio> File 0_map.situs - Header information:
lib_vio> Columns, rows, and sections: x=1-9, y=1-10, z=1-14
lib_vio> 3D coordinates of first voxel:
(396.000000,252.000000,132.000000)
lib_vio> Voxel size in Angstrom: 6.000000
lib_vio> Reading density data...
lib_vio> Volumetric data read from file 0_map.situs
lib_vwk> Setting density values below 0.000000 to zero.
lib_vwk> Remaining occupied volume: 1260 voxels.
lib_vwk> Map size changed from 9 x 10 x 14 to 9 x 11 x 15.
lib_vwk> New map origin (coord of first voxel):
(396.000,252.000,132.000)
lib_vwk> Map density info: max 54.000000, min 0.000000, ave
9.158249, sig 11.163736.
_____________________________________________________________________________
colacor> Processing atomic structure.
lib_pio> 2638 atoms read.
colacor> Geometric center: 420.843 279.341 169.848, radius: 77.497
Angstrom
_____________________________________________________________________________
lib_vwk> Generating Gaussian kernel with 5^3 = 125 voxels.
lib_vwk> Generating Gaussian kernel with 7^3 = 343 voxels.
lib_vwk> Generating kernel with 5^3 = 125 voxels.
lib_vwk> Map size expanded from 9 x 11 x 15 to 13 x 15 x 19 by
zero-padding.
lib_vwk> New map origin (coord of first voxel):
(384.000,240.000,120.000)
colacor> Projecting probe structure to lattice...
colacor> Computing fraction of PDB contained within the map (above
cutoff density) ...
colacor> Overlap fraction: 9.9460801E-01
colacor> Applying filters to target and probe maps...
colacor> Normalizing target and probe maps...
colacor> Target and probe maps:
lib_vwk> Map density info: max 5.378601, min 0.000000, ave 0.365617,
sig 0.871074.
lib_vwk> Map density info: max 5.313339, min 0.000000, ave 0.330269,
sig 0.909998.
colacor> Computing correlation value ...
colacor> Correlation value: 9.4922981E-01
_____________________________________________________________________________
colacor> Identifying inside or buried voxels...
colacor> Found 270 inside or buried voxels (out of a total of 3705).
colacor> Powell's optimization method.
colacor> Determining most efficient correlation algorithm based on
convergence and time...
colacor> Original algorithm: Correlation =
0.93791031 Time = 807.000000 us
colacor> Masked algorithm: Correlation
= 0.82471724 Time = 795.500000 us
colacor> One-step algorithm: Correlation =
0.93791032 Time = 5.436500 ms
colacor> Using original three-step correlation function.
colacor> Shown are: deviations (in A) from start structure,
colacor> Euler angles (in degrees), and correlation value.
colacor>
colacor> Performing optimizations...
colacor>
colacor> Powell optimization
colacor> X
Y
Z Psi
Theta Phi Correlation
colacor> 0.000 0.000
0.000 0.000 0.000
0.000 9.4922981E-01 Initial
colacor> 0.425 -0.438 -0.422
-10.258 5.386 0.993
9.5225197E-01 1
colacor> 0.418 -0.507 -0.636
-15.518 6.478 5.897
9.5248743E-01 2
colacor> 0.437 -0.524 -0.677
-21.248 6.820 8.884
9.5269450E-01 3
colacor> 0.411 -0.548 -0.687
-27.568 7.001 12.879 9.5291768E-01 4
colacor> 0.425 -0.531 -0.688
-28.638 7.109 14.098 9.5294922E-01 5
colacor> 0.477 -0.537 -0.713
-31.503 6.887 16.968 9.5299511E-01 6
colacor> 0.498 -0.522 -0.706
-36.830 6.521 22.348 9.5302053E-01 7
colacor> 0.480 -0.525 -0.712
-37.258 6.497 22.785 9.5302234E-01 8
colacor> 0.491 -0.506 -0.718
-38.535 6.611 24.020 9.5302311E-01 9
colacor> 0.491 -0.506 -0.718
321.465 6.611 24.020 9.5302311E-01
Final
colacor>
colacor> Powell optimization time: 2.646759 s
_____________________________________________________________________________
colacor> Writing result to file col_best_001.pdb.
_____________________________________________________________________________
colacor> All done!!!
|
The
program output shows how the
initial correlation of 0.949 increases slightly to 0.953 while the
structure is
matched. The output is written to the file col_best_001.pdb. Let's
rename
this file to "2_docked.pdb" to avoid overwriting it later.
Now
repeat this refinement also for the other files, 1_shift.pdb, and 1_rot.pdb. You
will notice that the initial correlation of these structures is much
lower.
The
correlation is the only criterion employed by colacor. In some cases, e.g. low
resolution or significant buried surface, the correlation criterion is
ambiguous or breaks down. In such cases the initial alignment "by eye"
based on structural expertise may become unstable and colacor may actually worsen the fit!
The user should experiment in such challenging cases with various
parameters of the program, and with turning the Laplacian filter on or
off (-corr option).
When
inspected
with VMD as above the results of this run should look similar to the
following
figure. In this case the shifted and rotated structures were
still within the "basin of attraction" of the correlation maximum, so
we do get good convergence of the results:
(Click image to
enlarge)
|
| Return
to the front page. |
|
