Dock uses several types of scoring functions to discriminate among orientations and molecules. Scoring is requested using the score_ligand parameter. Each scoring function is treated independently during the calculation and results are written to separate output files. In order to combine the results of two or more scoring functions, apply the additional functions in separate post-docking scoring calculations.
Dock will score the interactions of a ligand with the receptor if the intermolecular_score parameter is set. If flexible_ligand is set, then dock will score the interactions between rigid segments within a molecule if the intramolecular_score parameter is set. The total score is the sum of the intramolecular score and the intermolecular score, EXCEPT when different molecules are being compared for a rank_ligands list. After a flexible molecule has been docked, and it is being considered for the ranked ligand list, then the total score is set equal to only the intermolecular score plus whatever size penalty the user has specified with the contact_size_penalty parameter and so on for each type of scoring.
To enable rapid score evaluation during docking, the score potentials are precalculated on a three-dimensial grid using grid . However, continuum scoring may be performed by turning off the gridded_score flag. Continuum scoring may be used to evaluate a ligand:receptor complex without the investment of a grid calculation, or to perform a more detailed calculation without the numerical approximation of the grid. Continuum scoring is also triggered when an intramolecular_score is requested, but is only used for intramolecular score terms. When continuum scoring is requested, then score parameters normally supplied to grid , must also be supplied to dock. It is left to the user to make sure consistent values are supplied to both programs. Older grids calculated by chemgrid may also be read by specifying a value of 3.5 for the grid_version parameter.
The score is used to identify interesting orientations and molecules. If a top-scoring list is requested, like rank_orientations or rank_ligands , then dock will maintain a sorted list of a user-defined length for output. But if sorted lists are not requested, then dock will need to know what score cutoff to use to write out orientations or molecules. This cutoff is supplied by the user with the contact_maximum parameter and so on for each scoring function. This score cutoff may be overridden for orientations near the input orientation using the rmsd_override parameter.
Orientations may be filtered prior to scoring to discard those in which the molecule significantly overlaps receptor atoms. This feature is enabled with the bump_filter flag, but is only available if the gridded_score flag is also set. At the time of construction of the bump filter, the amount of atom VDW overlap is defined with the bump_overlap parameter. At the time of bump evaluation the number of allowed bumps is defined with the bump_maximum parameter. If Score Optimization is being performed, then a few number of bumps should be allowed, since the minimizer can recover from such clashes. In addition, a few bumps often indicate an orientation that interacts intimately with the site and often leads to a strongly favorable orientation after minimization.
The contact score function is enabled with the contact_score flag. The contact score is a simple summation of the number of heavy atom contacts between the ligand and receptor. At the time of construction of the contact scoring grid, the distance threshold defining a contact is set with the contact_cutoff_distance . Atom VDW overlaps are penalized by checking the bump filter grid, or with the contact_clash_overlap parameter for the intramolecular score. The amount of penalty is specified with the contact_clash_penalty parameter.
The contact score provides a simple assessment of shape complementarity. It can be useful for evaluating primarily non-polar interactions.
The energy score is activated with the energy_score parameter. It is based on the non-bonded terms of the molecular mechanic force field (please refer to Equation 1 on page 74 for more background). During grid construction (and continuum scoring) every term in the function may be tailored by the user. The distance dependence of the Coulombic function is set with the distance_dielectric parameter. The dielectric constant is adjusted with the dielectric_factor parameter. The distance dependence of the Lennard-Jones function is set with the attractive_exponent and repulsive_exponent parameters. Typically a 6-12 potential is used, but it can be softened up by using a 6-8 or 6-9 potential. Regardless of the exponent values selected, the same radii and well-depths are used. The VDW well-depths and radii are stored in an editable file (see vdw.defn on page 105 ) which is identified with the vdw_definition_file parameter. In addition, the model for non-polar hydrogens may be selected with the atom_model parameter. With a united-atom model, the non-polar hydrogens are given zero VDW potentials and any partial charge residing on them is transferred to the adjacent carbon. The united atom model provides a smoother intermolecular potential which requires fewer steps of minimization. However, the all-atom model is more accurate, and perhaps captures some aromatic interaction coulombic terms otherwise missing.
Chemical scoring allows the energy scoring function to be further tailored to enhance recognition of chemical complementarity. The attractive portion of the VDW term can often dominate the energy for uncharged molecules. With chemical scoring this term is scaled depending on the chemical labels assigned to the interacting atoms. It is activated by the chemical_score parameter. The chemical labels and definitions are the same as those for chemical matching (see chem.defn on page 106 ). The chemical interaction table resides in an editable file (see chem_score.tbl on page 108 ) and is identified with the chemical_score_file parameter.
Chemical scoring can be used to incorporate qualitative aspects of solvation. For instance hydrophobic-polar interactions can be made non-attractive or even repulsive. Further, it can be used to screen for molecules that contain a particular functional group (in concert with Chemical Matching ) for presentation to a receptor active site. The interaction table could also be derived using statistical techniques from binding and structure data to improve the modeling of a particular site or class of sites.
Chemical scoring is used for intermolecular scoring only. If intramolecular score is requested, then the regular energy score is computed for internal energy.
This type of scoring should be considered experimental. Parameterization is left to the user. It should be used at your own risk.
The RMSD score evaluates of the difference in conformation and orientation of two identical molecules. It is available when gridded_score is not requested and is activated by the rmsd_score parameter. The reference molecule is supplied with the receptor_atom_file , and the molecule to check is supplied with the ligand_atom_file , which may contain multiple molecules.
Since the RMSD is treated as a score, it may in fact be minimized. This procedure is useful for evaluating the difference of a CONCORD conformation from crystal conformation.
Score optimization allows the conformation and orientation of a molecule to be adjusted to improve the score. Although the calculation is expensive, it makes the conformation and orientation search more efficient because less sampling becomes necessary. Optimization is activated with the minimize_ligand parameter. The optimizer currently uses the simplex algorithm which does not require evaluation of derivatives. It does however depend on a random number generator which makes it not only sensitive to the initial seed provided with random_seed parameter, but also to the order of evaluation. So results will vary if molecules are supplied in a different order. The amount of variance should be small, though. For detailed calculations, it is recommended that the optimization be repeated with different random number seeds to check convergence.
The initial step size of the minimizer is specified with the initial_translation , initial_rotation , and initial_torsion parameters. The length of minimization may be controlled with the maximum_iterations parameter.
Even if several scoring functions have been requested, not all need to be minimized. Specification of which functions are to be minimized is done with contact_minimize and so forth for each scoring function. The termination criteria of minimization is specified with contact_convergence and so forth.
Since minimization may converge prematurely, each call to the minimizer is actually composed of multiple cycles. The number of cycles is controlled with the maximum_cycles parameter. Additional cycles of minimization are spawned if the previous simplex has made a significant difference in the conformation or orientation AND if the score has passed below a threshold set by the user. The difference in configuration is measured by the vector magnitude of the final simplex vertex array. The difference must exceed the cycle_convergence parameter to be significant. A value of 1.0 for this parameter would correspond to at least one of the simplex vertices moving a distance equal to the initial step size. The score threshold prevents repeated cycles of minimization of a configuration that is really of no interest. The score threshold is set with contact_termination an so forth for each scoring function.
The performance of the minimizer can be monitored using the -p flag (see Command-line Arguments on page 53 and Performance on page 73 ).