Secondary and Tertiary Structure Prediction of Proteins: A Bioinformatic Approach - Complex System Modelling and Control Through Intelligent Soft Computations - page 543

In comparative modeling, target sequences often have few inserted residues as
compared to the template structures. Thus, no structural information about these
inserted regions could be obtained from the template structures. These regions are
called surface loops. Loops often play an important role in de

ning the functional

speci city of a given protein structure, forming the active and binding sites for drug

molecules. The accuracy of loop modeling is a major issue for comparative models
for applications such as protein-ligand docking i.e. structure based drug design.
There are two main classes of loop modeling methods:

(a) Database search approaches, where a small loop of 3 10 amino acid residues

are searched in a database of known protein structures and if such loops

-

t the
criteria of lowest energy, such loops are selected and added to the model
structure. All major molecular modeling programs and servers support this
approach e.g. Modeller ( Š ali and Blundell 1993 ), Swiss-Model (Guex and

Peitsch 1997 ).
(b) The conformational search approaches mainly depend on an ef

cient energy
function to choose the loop with lowest energy. If required, energy of the
selected loop is minimized using Monte Carlo or molecular dynamics simu-
lations by AMBER, and GROMACS techniques in order to arrive at the best
loop conformation with lowest energy.
Side chain modeling is also one of the essential components in structure pre-
diction of proteins. When we compare the side-chain conformations (rotamers) of
residues that are conserved in structurally similar proteins, we copy coordinates of
conserved amino acid residues entirely from the template to the model. But when
we have different residues, side chains are added to each amino acid and their all
possible rotamers are searched to find the most stable (having least energy) rotamer

from rotamer library.

3.3.4 Model Re nement

One of the major limitations of computational protein structure prediction is the
deviation of predicted models from their experimentally derived true, native
structures. Re nement of the protein model is required, if there is problem in

structural packing of side chains, loops, and secondary structural elements in the
target model. For any error in backbone or side chain packing, energy minimization
is done which requires an enormous precision in the energy function. At every
minimization step, a few big errors (like bumps, i.e., too short atomic distances) are
removed while many small errors might be introduced which lead to another dis-
tortion in the structure. In energy minimization, force fields must be fast to handle

these large molecules ef nement of the low resolution predicted models

to high resolution structures are close to the native state, however, it has proven to
be extremely challenging. There are various programmes e.g. GROMACS ( http://

ciently. Re