DENSSWeb runs a front end interface to DENSS and uses EMAN2 to perform
alignment and averaging of 20 independent reconstructions.
DENSS works best with a smooth fit to the SAXS profile instead of using the raw
data. To get a smooth fit, use an indirect
Fourier transform algorithm such as GNOM. These methods
take advantage of the oversampling provided by solution scattering experiments.
The input for DENSS is either the GNOM .out file or a 3 column ASCII text
file where the columns are q, Intensity, error. Make sure there’s no zeros
anywhere in the file (including the errors columns and first few rows of
the file, which sometimes are zeros due to the beamstop).
Each job submitted to DENSSWeb will be given a unique ID code. However, the
Job Name field can be used for giving your job a more descriptive name.
This field can only contain letters, numbers, dashes (-) and underscores
(_) (no spaces or special characters allowed).
If using a standard .dat file (3-column ASCII text), you must also enter
the estimated maximum dimension of the particle (this partly defines the
size of the box containing the particle). When using a GNOM file, the
estimated maximum dimension (Dmax) is taken directly from the .out file and
does not need to be entered in the input field. However, you may enter a
different Dmax than in the .out file if you like.
To keep DENSSWeb running efficiently and allowing multiple users to use the
server, DENSSWeb always sets the number of samples, N, to be 32 and does
not allow the user to change this (computation scales as N^3).
While N is set to 32, one can change the oversampling from the default of
3.0. The real space voxel size will adjust to ensure N=32. Oversampling as
low as 2 can work in some simple cases. However, for complex cases larger
oversampling (say 5), may be needed to ensure that the calculated
scattering matches the data. If the fits look poor (*_fit.png or
*_map.fit), increase oversampling to try and improve the number of grid
points that get scaled in reciprocal space. Note that this will increase
the real space voxel size, lowering resolution. However, in most cases the
reconstructions do not support such high resolutions, so it shouldn’t be a
problem. To see if it may be a problem, check out the FSC curve in the
folder spt_01/fsc_0.txt, if FSC falls below 0.5 before the end of the
x-axis, you’re unlikely to gain resolution by decreasing the voxel size
anyways. If you want to increase the oversampling *and* decrease the voxel
size (thus requiring N>32), you must download DENSS and EMAN2 and run
things yourself. This is made simpler with the superdenss bash script
available on the DENSS github page here.
If you'd like to set the actual number of electrons in the map to be
something you know, for example from the molecular weight of the particle,
enter the number of electrons in the Electrons input. This simply scales
the density of the final map such that the total number of electrons
matches the input value.
Enter the reCaptcha number to ensure you're not a robot and click Submit.
Typical times are about 5 minutes per job.
When complete, the browser will display an embedded LiteMol window showing
the electron density map. This display is interactive, including changing
the isolevels (click the tools icon in the LiteMol window).
Below the interactive density display is a plot of the Fourier Shell
Correlation as a function of resolution. The estimated resolution of the
averaged reconstruction is given by where the FSC falls below 0.5, and that
value is displayed on the plot.
Below the FSC curve is a series of three plots showing various statistics
(chi^2, Rg, Support Volume) as a function of step number. Plots for all 20
reconstructions are shown.