Outstanding efficiency of ligand docking and binding energy estimations achieved by Lead-Finder are due to unique docking algorithm and extra precision representation of protein-ligand interactions.
From mathematical point of view ligand docking represents global minimum search on the
multidimensional surface describing the free energy of protein-ligand binding. With ligands having
up to 20-30 degrees of freedom (freely rotatable bonds) and complex nature of energy
surface, global optimum search represents generally unsolved scientific task. To tackle this
computationally challenging problem Lead-Finder applies unique approach combining genetic algorithm search,
local optimization procedures, smart exploitation of the knowledge generated during the search run.
Rational combination of different optimization strategies makes Lead-Finder efficient in terms
of coarse sampling of ligand?s phase space and refinement of promising solutions.
Entire docking algorithm has several tenths of settings, each of which has certain impact on the
algorithm performance, robustness and speed of calculations. These relations were studied
by us and for the sake of user comfort all settings were zipped in to two regimes
of ligand docking algorithm with carefully balanced speed/accuracy ratio. First one is the default
docking regime, which settings were adjusted to achieve maximum docking accuracy at a reasonable
time of calculations. Second is the so-called screening regime, which was designed
to be maximally fast at the cost of small (~5%) decrease in docking accuracy
(see section Benchmarks).
More details about the docking algorithm of Lead-Finder.
Extra precise representation of protein-ligand interactions implemented in Lead-Finder scoring
function is the second (in addition to docking algorithm) component of successful ligand docking.
Lead-Finder scoring function is based on a semi-empiric molecular mechanical functional, which explicitly
accounts for different types of molecular interactions. Individual energy contributions are scaled with empiric
coefficients to fit particular purposes: accurate binding energy predictions, correct energy-ranking of docked
ligand poses, correct rank-ordering of active and inactive compounds during virtual screening experiments.
For these reasons three distinct types of scoring functions based on the same set of energy contributions but
different sets of energy-scaling coefficients are used by Lead-Finder.
First type of scoring function is designed for accurate estimation of free energy
of ligand binding based on the structure of protein-ligand complex, and for this reason
is called ΔG-scoring function. Experimentally determined structure of protein-ligand complex,
or a modeled structure (obtained for example by docking or/and molecular dynamical studies)
can sever as input for binding energy estimation with ΔG-scoring function. Scaling coefficients for
this function were derived by fitting calculated binding energies to the experimental values for
the set of protein-ligand complexes with known 3D-structure and experimentally measured binding constants
(see section Accuracy of binding energy estimations).
Second type ranking scoring function serves for correct ranking of ligand
poses during the docking run. The purpose of this scoring function is to give highest score
to the correct (experimentally observed) ligand pose. Parameterization (choice of scaling
coefficients) of this scoring function aimed maximum docking success rate (maximum number
of top-scored poses with correct geometry) for the fixed set of docking-predicted ligand poses
over the set of protein-ligand complexes with known 3D-structures (see section Docking success rate).
Finally, special set of scaling coefficients was designed to yield maximum efficiency in virtual
screening experiments, that is to assign higher score to active ligands (true binders)
than to inactive ones. The latter scoring function type is called virtual screening
or VS-scoring function. Assessment of VS-scoring function is illustrated in section
Virtual Screening Performance.
More details about Lead-Finder scoring function.
|