Biomedical Engineering Reference
In-Depth Information
4.2.4 Relaxing the Structure
All-atom force-fields give rise to rugged energy landscapes. Subtle changes of
internal degrees of freedom can yield drastic fluctuations in energy, prohibiting
a direct evaluation of putative structures without relaxing the structure in the
respective force-field. The parameterisation of structure prediction force-fields
generally neglects any dynamics (and thus energy barriers), such that only the
local minima are informative with respect to the likelihood of the structure.
42,44
Hence structures have to be relaxed towards a nearby energy minimum for a
meaningful evaluation of their quality. It makes sense to view this relaxation as
an intrinsic part of the energy evaluation and hence we discuss suitable
methods here rather than in the section on optimisation methods. Moreover,
the complete process of an energy relaxation usually entails only a small
overall change of the protein backbone, which distinguishes relaxation further
from the broad sampling required to solve structure calculation problems.
An efficient relaxation is highly challenging due to the steepness of the
Lennard-Jones potential and the stiffness of the covalent parameters. Simple
strategies involve simulated annealing with very small torsional moves, as used
e.g., for relaxation in the structure-prediction force-field PFF02.
45
ROSETTA
interlaces dedicated side-chain rotameric sampling (called side-chain repack-
ing) with torsion-space minimisation.
26
During this process the structure is
compressed and expanded to 'shake loose' possibly interlocked side-chains.
The breathing effect is obtained by ramping between a soft-core and the full
Lennard-Jones potentials in multiple cycles. Backbone torsions are kept
constant during re-packing and only distinct rotameric states are sampled. This
enables enhanced computational efficiency by pre-computation of all
interaction energies followed by rapid simulated annealing within the discrete
rotameric configuration space.
46
4.2.5 ROSETTA All-Atom Force-Field Accuracy
If experimental restraints are sparse, multiple distinct conformations may
satisfy all experimental information. Convergence of the structure calculation
then hinges on the ability of the force-field to discriminate non-native (decoys)
from native structures. The accuracy of RAAFF has recently been tested on
111 protein domains using large sets of decoys without any experimental
restraints.
47
The test revealed a remarkably high fidelity: for 41% of the
proteins examined, the lowest-energy structure is within 1.2
˚
CaRMSD from
the deposited crystal structure, and for 70%, it is within 2.5
˚
CaRMSD
47
(Figure 4.3). Deviations were mainly found in solvent-exposed loop-regions
rather than the protein core and could be rationalised with the presence of
crystal contacts, ligands, or oligomeric binding partners omitted from the
calculation. Indeed most deviations above 1.5
˚
CaRMSD disappear when
models are computed in the context of the crystal lattice or multimer.
47
Despite the overall very positive result of the benchmark some caution with
its
interpretation
is
required.
For
many
of
the
111
protein
domains
an
