Chemistry Reference
In-Depth Information
algorithms (GlamCock [241], ICM [240]), genetic algorithms (Autodock [105,417], Gold [104], MolDock [245]),
hierarchical scoring functions for crude shape fitting and linear optimization of ligand pose (QXP [251], LigandFit
[252], Glide [112]) and systematic analysis of possible minima using graph searches (eHits [250].
Docking methods are often characterized by docking success rates which specifies the percentage of correctly
predicted ligand positions as well as virtual screening studies in which ligands known to be active for a particular
target protein are mixed with inactive ones (decoys) and rank-ordered according to their predicted scores. It is
important to make standard benchmarking properties which include test sets of protein-ligand complexes, data
sharing, preparation of protein ligand structures for modeling sets of active and decoy ligands for virtual screening
experiments. Although programs like AutoDock and Glide estimate the free energy of protein-ligand binding it is
believed that binding energy prediction is a post-docking procedure and therefore more computationally demanding
methods such as linear interaction energy approximations and free energy perturbation methods are required. We
note that one of the objectives is still to improve speed and quality of binding energies calculated from docking
programs [105,417, 112].
Three specialized high accuracy scoring functions and an innovative molecular docking algorithm were recently
introduced in the Lead Finder docking software [79]. This algorithm combines the classical genetic algorithm with
various local optimization procedures and resourceful exploitation of the knowledge generated during the docking
process. The scoring functions are based on a molecular mechanics functional which explicitly accounts for different
types of energy contributions scaled with empirical coefficients to produce three scoring functions tailored for 1)
accurate binding energy predictions 2) correct energy-ranking of docked ligands poses and 3) correct rank-ordering
of active and inactive compounds in virtual screening experiments.
The purpose of the first function is to accurately estimate the free energy of ligand binding for a particular structure
of a protein-protein complex. The purpose of the second scoring function is to give the highest score to the correct
experimental ligand pose. Finally, another set of scaling coefficients was designed to yield maximum efficiency in
virtual screening experiments. The local pose optimization is viewed as a valuable component of genetic algorithm
which facilitates faster evolution of individuals in addition to usual genetic operations such as recombination and
mutation. The pseudo Solis-Wets (PSW) optimization is based on a random displacement in each degree of freedom,
following the chosen direction when the energy of a new ligand pose is lower. Before the start of a docking process,
the conformation of a ligand in solution is optimized, its energy is then used as a reference in energy calculations
and its structure used as a source for generating the initial pool of individuals by randomizing the translational and
orientation coordinates of a molecule. The genetic algorithm uses the notion of a niche to cluster individuals with
similar genotypes and to restrict their expansion [79, 105].
The protein structure, in an aqueous environment for a living organism, can differ from the monocrystal structure.
3D structures determined by NMR spectroscopy are not completely sensitive to the conformational state of the
molecules and are less reliable than X-ray results, which does not always yield the structural determination of some
fragments due to the partial disorder in crystals. In addition, in order to reduce the time to obtain X-ray results, the
resolution is often chosen to be lower than necessary for reliable localization of the hydrogen atoms, yielding
imprecise determination of which tautomeric forms of ligand and proteins are interacting [98, 298].
In addition to the unreliability of some 3D structures of proteins used in docking programs, the scoring functions are
also not always examined for macroscopic physicochemical properties and consist in general of discrete values
dependent of the types of interactions, i.e hydrophobic, hydrophobic, H-acceptor, H-donor,
etc.
Some scoring
functions characterize Gibbs free energy or the potential energy to determine the thermodynamics of ligand-receptor
interactions. The question arises however, if this criteria is correct. Does ligand and receptor know whether their
interaction is stable or unstable, with low or high free energies respectively, or does the kinetic factors play a more
important role in the interaction? Is there a strong dependence on the interaction of some active centers of the
molecules and not on the total interaction energy, leading to the sometimes observed poor relation between scoring
functions? What about Gibbs free energies and other parameters describing the biological activities of the
compounds?
Another important issue is insufficient consideration of the receptor and ligand flexibilities. In some methods a ligand
is decomposed into small rigid fragments, which are reassembled to fit the binding site. The problem may not have an
