two reference methods and two new methods, respectively:
- random walk
- Metropolis Monte Carlo Simulated Annealing
- Stochastic Approximation with Smoothing (SAS)
- Terminal Repeller Unconstrained Subenergy Tunneling (TRUST)
Of particular interest is whether any of these algorithms could be used
to dock a database of
typical small molecules in a reasonable amount of time.
To address this question, each algorithm was used to dock four small
molecules presenting a
wide range of sizes, degrees of flexibility and types of interactions.
Of the algorithms tested
only stochastic approximation with smoothing appeared to be sufficiently
fast and reliable to
be useful for database searches. This algorithm can reliably dock relatively
small and fairly
rigid molecules in a few seconds and larger and more flexible molecules
in a few minutes.
The remaining algorithms tested were able to reliably dock the small and
fairly rigid molecules
but showed little or no reliability when docking large or flexible molecules.
Conceptually the SAS algorithm 'averages' the energy landscape by convoluting
it with a
Gaussian. As a result the transformed landscape only has one minimum, which
can easily be
found by conjugate gradient methods. The molecular docking mode at that
minimum is then
taken as a starting point for minimization in an energy landscape which
has been 'averaged' to
a lesser degree. This process is repeated until the 'averaging' is so negligable
that minimization
takes place in the original energy landscape. The method is related to
the Diffusion Equation
Method used in protein folding simulations but there are significant implementation
differences.
Reference:
Diller D.J., Verlinde, C.L..M.J. (1999). A critical evaluation of several global optimization algorithms for the purpose of molecular docking. J. Comp. Chem. 20, 1527-1532.