· Delarue web servers
· Nomad Flow-chart
· Normal Mode calculation
  · Examples (Movies)
  · Submit a job (from PDB file)
  · Split trajectory (for MR)
  · Generate decoys (MixMod)
  · Elastic Energy (Perturb. Anal.)
· Overlap coefficients
  · Submit a job
  · Include Profit
· X-Ray refinement
  · Standard refinement
  · Screening Mol. Repl. Solns.
· EM refinement
  · Documents/Examples
  · Get Structure Factors
  · Submit a job (no NCS)
  · Submit job with NCS
· Docking refinement
  · Submit a job
· Force field methods
  · Energy minimization
  · Gromacs NMA
· References
· Homes/Links
· Job queue status

Generate NMA decoys around a structure

Normal modes provides an efficient way to automatically generate rough decoys from protein structures, e.g. to train algorithms that try to discriminate between correct and incorrect structures. Since normal modes usually lead to somewhat distorted local structure (e.g. bonds, angles), you should energy minimize (regularize) the resulting decoys before using them for training.

The server here can generate arbitrary numbers of decoys,
-either by scanning along a single mode (in that case you should use the split traj. option)
-or by (randomly mixing and) combining amplitudes from multiple (user-specified) modes to reach a desired cRMS value, specified by the user below.

Choose the number of modes you want.
The scale (amplitude) of the movement can be chosen by controlling the average rmsd of the trajectory. The rmsd per residue is given on output for each chosen mode (as a plot).

Your email adress: (Recommended, for notification)

Job title: (Only alphanumerical characters)

PDB file to calculate modes for:

First normal mode to use for decoy generation:
(1--6 are translation and rotation.)

Last normal mode to use for decoy generation:

Distance weight parameter (Ångström):

Cutoff to use for mode calculation (Ångström):

Number of decoys to generate:

Approximate average cRMS (Ångström) of decoys:

Method to use (advanced):

Input data formats

  • The job title is just for your own identification, but note that it will show up in the public job queue (but your results will not be public).

  • The coordinate file should be in PDB format, with only a single structure (no multiple models). Atoms marked with alternate residue flags will be removed. Whatever atoms are in the file will be used for the calculation. The length of each mode vector will be 3*natoms. It is better not to include Hydrogen atoms in the PDB file.

  • All interactions are weighted by exp(-(r/r0)^2), where r0 is a distance-weight parameter. Higher values lead to a stiffer system, while lower values decouple local and global motions. For an example, consider all-atom modes in a protein. A low value of r0 will create larger hydrogen motions relative to the heavy atoms, even for low frequency eigenmodes. A value of 3.0 Angstrom works well for CA-only models, but you might prefer a larger value (5.0-10.0 Angstrom) for all-atom structures. See the Hinsen article in the References section.

  • A cutoff is used in the mode calculation. In the Tirion model (Elastic Network Model) only those pairs of atoms that are closer than the cutoff are linked by a spring of universal length. Ideally this should be choosen so that the weighting causes the interactions to be negligible outside the cutoff, but in practice a cutoff around 10 Angstrom works fine in almost all cases. In general you might be able to use a smaller cutoff value for all-atom calculations than CA-only calculation. See the Tirion article in the References section.

  • You can choose how many modes to calculate. The execution time for the sparse matrix solver is more or less proportional to the number of modes requested, while the full matrix solver execution is dominated by the reduction to tridiagonal form - the actual mode calculation is very fast in that case. Due to the memory requirements you cannot use the full solver for more than about 5000 atoms, and the sparse solver is much faster even for small systems (from about 500 atoms) if you only want the first (lowest-frequency) 10-100 modes. Remember that the first six modes correspond to rigid body motions (translation and rotation), so the structurally interesting ones begin at 7.

  • If you really want to you can override the automatic choice and specify either the sparse or full solver. Be aware that the full solver wont work for more than 5000 atoms, but it is faster for small systems or if you want a large number of mode vectors (or if you wanted a huge cutoff for some reason).

  • Some timing examples for 1ANF with 2860 atoms (Hessian dimension is 8580 by 8580): Calculating 20 modes takes 33 seconds and for 100 modes it increases to 1 minute and 40 seconds.

  Marc Delarue http://lorentz.dynstr.pasteur.fr
Page last modified 17:47 September 22, 2009.