Deformable Elastic Network (DEN)
At low resolution the number of experimental observables is usually
smaller than the number of parameters. Overfitting thus becomes
a major issue.
The main of the DEN approach is to refine only those degrees of freedom
for wich the data provide information.
For those degrees of freedom that are
not defined by the data we use prior structural information.
Such prior information could e.g. come from a know X-ray structure in a different
conformation or a homology model. This model is referred to here as the 'reference model'.
To balance prior information and data (density map) an elastic network is defined
on the reference model, i.e. a number of harmonic restraints are defined between
randomly chosen pairs of atoms that are within a distance range of typically 3 to 15 Å.
During the refinement the elastic network is allowed to deform, i.e. the equilibrium
distances can (slowly) follow any changes during the refinement. This way the minimum
of the elastic network potential moves. However, the equilibrium distances are
attached by springs to the corresponding reference distances. The strength of these
springs are defined by a parameter γ. A small γ-value keeps the structure
close to the reference model and a large γ-value allows largers structural
changes.
The figure below demonstrates the principle of the DEN method: