Biomedical Engineering Reference
In-Depth Information
Figure 4.3
Accuracy of the ROSETTA all-atom force-field (RAAFF) in a structure-
prediction benchmark of 111 protein targets. Each panel represents a
different protein target. The y-axis is the ROSETTA all-atom energy and
the x-axis is the CaRMSD from the crystal structure; red dots are models
relaxed from the crystal structure. The inset shows the energy landscape
for 1TEN (a fibronectin type-III domain) in more detail and a
superposition of the models within four energy units of the lowest-energy
model (indicated by the horizontal grey line in the plot) on the crystal
structure (black). Colours indicate amount of variation in the ROSETTA
ensemble (blue, low; red, high); variation is concentrated toward the
loops. The vertical grey bars indicate the 1 and 2
˚
points. For 41% of the
proteins examined, the lowest-energy structure is within 1.2
˚
CaRMSD
from the deposited crystal structure (as for 1TEN), and for 70%, it is
within 2.5
˚
CaRMSD (A-D). Reproduced from ref. 47 with permission,
# Elsevier, 2011.
unguided standard ROSETTA de novo structure prediction would not have
been able to sample sufficiently well around the native structure to reveal its
distinct native energy funnel. To map the entire energy landscape despite this
limitation, near-native sampling had to be enhanced artificially using
information from the native structure. The lack of a sufficiently thorough
sampling method that can find the low-energy region of the native state
without bias, raises the possibility that there are other low-energy regions
which cannot currently be detected.
In summary, the 111-protein benchmark shows that the energy function is
highly accurate, but without sufficient near-native sampling we cannot benefit
from this in structure calculation. Systematic force-field artefacts could not be
