The Many Faces of Structure-Based Potentials: From Protein Folding Landscapes to Structural Characterization of Complex Biomolecules - Computational Modeling of Biological Systems - page 49

4.4 Molecular Modeling

SBM are structurally robust, which makes them ideal candidates for molecular
modeling applications. During molecular dynamics the native bias maintains a
native-like configuration but all interactions are malleable. Under molecular dynam-
ics, a system populates the lowest free energy basins, and coupled with simulated
annealing can even search for the lowest potential energy minima [ 63 ]. Through the
introduction of external biasing potentials, AA SBMs built from high-resolution
structures can reveal candidate AA structures from low resolution experimental
data.
In a recent study of the ribosomal elongation cycle, Ratje et al. [ 50 ]used
multiparticle cryoelectron microscopy analysis to capture subpopulations of EF-G-
ribosome complexes at subnanometer resolution. While this resolution is not fine
enough to achieve atomic details, the known crystallographic structure can be used
to obtain atomic models of the microscopy data with a procedure termed MDFIT
[ 65 ]. MDFIT biases the AA SBM with an energetic term developed in Orzechowski
and Tama [ 48 ], which uses the correlation between the simulated and experimental
electron density. The overall potential function therefore becomes
V model D V AA C V map D V AA C W X
ij k

sim

ij k exp

ijk ;

(13)

and exp is an overall weight and sim

ij k

where W

ij k are the normalized electron
densities at voxel .i;j;k/and V AA is the AA SBM potential. A molecular dynamics
simulation initialized at the crystallographic structure will distort to maximize the
overlap between the simulated structure and the experimental electron density. The
structure-based potential naturally maintains tertiary contacts present in the crystal
structure without the need for ad hoc restraints.
The electron density map works well as a global bias, but local biases can also be
introduced. Candidate structures for protein-protein complexes can be derived by
introducing interprotein contacts from bioinformatic analysis and minimizing the
resulting structure-based potential with molecular dynamics. Schug et al. [ 51 ]were
able to predict the structure of the Spo0B/Spo0F two-component signal transduction
(TCS) complex within 2.5 A of an existing crystal structure. TCS is ruled by
transient interactions, posing harsh challenges to obtain atomic resolution structures.
These transient interactions though have bioinformatic signatures, which provide
the external biasing potential needed for modeling. Short-range contact potentials
were introduced between correlated residues and the resulting potential

X

V model D V AA C k.r CM / 2 C C AA .r ij ; r/; (14)

f i;j g