Biomedical Engineering Reference
In-Depth Information
refinement in ROSETTA
24
improve the structural quality of NMR solution
structures. The intuitive notion that better force-fields improve structure
determination when only sparse data is available was recently justified using
theoretical means.
32
Habeck compared a flat prior,aprior whose non-bonded
forces are only repulsive and a prior with attractive and repulsive non-bonded
terms. The most informative prior not only increased the posterior density for
the
native
structure
but
also
was
supported
most
strongly
by
the
data
according to its Bayes factor.
32
All-atom force-fields used for structure prediction render structure
determination sampling-limited rather than information-limited. The
ROSETTA all-atom force-field (RAAFF), for instance, is informative and
accurate, but difficult to sample due to its ruggedness and the dominance of
short-range interactions (Section 4.2). Using RAAFF and sparse backbone-
only NMR data, we found that addition of sparse NMR restraints actually
resulted in sampling of lower RAAFF energies.
26
Instead of just increasing the
precision of the calculation by restricting to restraint-compliant conformations
the data served as a guide towards the sharp and informative native energy
basin. This basin proved to be sufficiently narrow to yield precise and accurate
structures of atomic resolution without further need of experimental data.
Thus, the calculations using RAAFF were sampling-limited. Non-informative
priors, in contrast, are easy to sample but require the experimental data to yield
precise information (information-limited).
It follows from this discussion that the characteristics of structure
calculations and thus its challenges change dramatically when a structure
prediction force-field is used. Less data is required to achieve high resolution,
but adequate sampling of conformational space is difficult to achieve. To
account for this dramatic change in the characteristics of structure calculation
we introduce instructiveness and resolution as two descriptors for the behaviour
of
experimental
restraints
in
the
context
of
sampling-limited
structure
calculations.
The instructiveness and resolution of a restraint, are determined by the shape
of the restraint-energy landscape distant from and close to the native structure,
respectively (Figure 4.1). An instructive restraint yields a strong energy
gradient far away from the native structure that is able to guide the
conformational search. The resolution of a restraint set is determined by the
broadness of the native energy basin and corresponds to the amount of
structural variation that is not penalised significantly by the restraints. This
distinction between instructiveness and resolution of restraints also leads to
distinct definitions of sparseness: a data set may be called guidance-sparse if the
available instructive data (e.g., NOE data) is not sufficient to guide the
conformational search to the lowest-energy region (Figure 4.2). A data setis
information-sparse if the resolution of the data is not sufficient to reach the
required precision.
Means to overcome information sparseness are additional experimental data
and high-resolution force-fields. To overcome the lack of guidance one can
