Computational Approaches in Peptide and Protein Design: An Overview - Peptide and Protein Design for Biopharmaceutical Applications - page 25

ligand within the binding site of the receptor. In this case, however, para-
meters for sampling may be not only the dihedral angles of the Cartesian
atom coordinates of the ligand, but also 'global' parameters determining
orientation of the ligand with respect to the receptor as a whole. On the
other hand, there are sampling methods summarily known as genetic
algorithms . Unlike Monte Carlo or molecular dynamics simulations, this
approach starts with an initial population of different conformations and/
or orientations (configurations) of the ligand. Each configuration is defined
by a set of parameters, both global and ligand-conformational ones, which
are treated as sets of genes in a chromosome, i.e. each may experience
'mutations', 'crossovers' and 'migrations' by analogy with genetic pro-
cesses. Acceptance or rejection of the next population of configurations is
governed by values of the corresponding scoring functions. The process of
sampling stops when the value of the scoring function has converged,
usually after several hundreds of steps of 'genetic perturbations', yielding
an ensemble of plausible configurations (poses) of the ligand in the binding
site. Genetic algorithms are available in a variety of molecular modelling
programs, such as the ligand-receptor docking program AutoDock, which
still is the single most widely employed docking program [34].
Sampling of protein conformational space The problem of sampling
conformational space for proteins has, in fact, two very different limitations.
On the one hand, the number of parameters (dimensions) defining confor-
mational space in proteins is much larger than in peptides, which makes any
of the sampling protocols outlined above both more complicated and more
time consuming. But on the other hand, most proteins, unlike peptides, exist
in a single native well-stabilized 3D structure of main interest in most
practical applications of protein design. Historically, procedures for sam-
pling conformational space in proteins were developed mostly in attempts to
understand possible mechanisms of protein folding, a very general problem
of molecular biology. However, as noted recently by Dill [61], protein
folding is, in fact, three somewhat different problems. The first is the
computational problem of predicting the stable 3D structure of the protein
from its amino acid sequence (protein-fold prediction), and the other two
relate to the thermodynamics of folding (how the native structure results
from interatomic forces acting on the amino acid sequence) and to its
kinetics (how the protein achieves the native fold from non-native ones).
The first problem relates to equilibrium and is path-independent and there-
fore much closer to solution than the latter two. It is also the most relevant to
the sampling problem in computational protein design. Some approaches
used for protein design are briefly outlined below.