Kuntz Home / DOCK Home / DOCKumentation Contents / New Features
prev section prev toc up next next section

Force-Field Score Optimization

Daniel Gschwend
Overview / Usage / Cautions

Overview

DOCK 3.5 now has the ability to refine each orientation generated prior to scoring it (Gschwend, manuscript in preparation). The refinement takes the form of a rigid-body minimization of force-field score. Note that the minimizer is optimizing the non-bonded interactions between ligand and receptor; torsions, bond lengths, and bond angles are not modified. The refinement of ligand interactions is crucial for retrieving orientations which would otherwise be thrown out due to short-range steric clashes. Known crystal complexes can be salvaged with very minimal sampling when minimization while docking is employed. Furthermore, more representative interaction scores are obtained without the need for post-docking refinement.

A grid-based simplex minimizer is used to achieve a speed of one to fiveminimizations per second (SGI Indigo R3000). The simplex supplants a quasi-Newton DFP minimizer because it is more than twice as fast and produces results of improved quality. The quasi-Newton is very sensitive to local curvature and can easily get stuck on the very jagged grid-based potential surface. The simplex also requires no derivatives. One change has been made to the van der Waals parameter file: polar hydrogens are no longer volumeless. Polar hydrogens were originally treated as being volumeless to allow for closer heteroatom contacts in the form of pseudo-hydrogen-bonds. The minimizer is clever enough to realize that nothing prevents the approach of one charged atom to the nucleus of a polar hydrogen, resulting in skyrocketing electrostatic interactions. Hence, polar hydrogens have been given a small (0.6Å) but non-zero van der Waals radius. This change is reflected in vdw.parms.amb.mindock, the parameter file which should be used for all DOCK runs employing minimization.

Usage

  1. Because optimization is currently supported only for force-field scores and not contact scores, it is recommended that the "force-field scoring only" scoring option (4) be used. This will require the use of chemgrid for generation of scoring grids.

    When force-field scoring and contact scoring are used in conjunction (scoring option 3), the contact score determines whether the orientation should be minimized, not the number of bumps. Beware that when both a contact score and force-field score are given in the output, the force-field score corresponds to the minimized energy for an orientation, whereas the contact score corresponds to the unminimized orientation, as contact scores are not re-evaluated after optimization.

  2. Add the minimize keyword to your INDOCK parameter file.

  3. Change the name of the van der Waals parameter file to vdw.parms.amb.mindock.

  4. Increase the number of bumps allowed in docking. The number of allowed bumps (keyword bump_maximum) should be non-zero, since minimization will resolve steric clashes. Start with at least 2 bumps and allow more for larger ligands. (A more powerful method is to increase the active site volume, but run-times will increase. Please consult the documentation for rebump.)

  5. Decrease the sampling level by reducing bin sizes. Sampling may be greatly reduced when using minimization because of the effectiveness of the matching algorithm. It is recommended that non-zero overlaps be used. Run-times can become severely inflated with larger bin sizes when minimization is performed.

For information on speeding up the process of minimization by judiciously selecting which orientations to minimize, please see the section entitled Degeneracy Checking.

Cautions

Minimizer randomness

The first step in each minimization is of random magnitude in each of the six dimensions. Random numbers are generated based on a "seed" value which dictates a particular sequence of random numbers. Each time a random number is requested, the next number off this list is taken. Currently, the seed value remains a constant, so that identical DOCK runs performed sequentially will in fact produce identical results. However, should the order of the ligands and/or orientations be shuffled, different results will be obtained due to the altered sequence in which random numbers are obtained for each ligand. Furthermore, if one performs two sequential DOCK runs with altered parameters such that the number of orientations minimized changes (e.g. increasing the number of bumps), the random number corresponding orientations receive will differ. Again, different, perhaps unexpected, results are obtained, as the results of the first DOCK run will no longer be a subset of the second set of results (in the example if increasing the number of allowed bumps).
prev section prev toc up next next section
Curator: Daniel Gschwend, gschwend@cgl.ucsf.edu (rev. 1 September 1995)