HYBRID docks multiconformer molecules into one or more receptors with bound ligands using an exhaustive search that systematically searches rotations and translations of each conformer of the ligand within the active site. Following the exhaustive search the top scoring poses are optimized and assigned a final score. These two steps are described in more detail below
If there is only one receptor this these steps are skipped and all ligands are docked to the receptor.
- For each ligand in the docking database the similarity of the ligand to each of the receptor’s bound ligands is calculated based on the 3D shape and chemical similarity of the docking ligand and bound ligand.
- Each ligand is then docked to the single receptor that has the most similar bound ligand as described in the following steps.
- Enumerates, to given resolution, every possible rotation and translation of each conformer of the ligand being docked within a box enclosing the active site. The resolution of the exhaustive search is determined by the overall resolution setting of -dock_resolution flag.
- Discard poses that either clash with the protein or extend to far from the binding site using the receptor’s negative image outer contour (see Negative Image section).
- If the negative image inner contour is enabled discard any poses that do not have at least one heavy atom that falls within the inner contour. (see Negative Image section).
- Discard any poses that do no match any user specified constraints. (see Constraints section).
- Score all remaining poses using Chemical Gaussian Overlay.
- Sort poses by score and pass the top scoring poses to optimization.
- Enumerate nearby positions of each pose by having the initial pose take one positive and one negative step for each translational and rotational degree of freedom (729 poses total). The resolution of these steps is half that of the exhaustive search, and is determined by the setting of the -dock_resolution flag.
- Each pose is scored with Chemgauss4.
- The best scoring poses are retained. The number of poses retained is determined by the setting of the -num_poses parameter.
- The overall score of the molecule, used to rank the molecule against other molecules in the docking database, is set to the best scoring pose’s score.
The Chemgauss3 scoring function uses Gaussian smoothed potentials to measure the complementarity of ligand poses within the active site. Chemgauss3 recognizes the following types of interactions.
- Hydrogen bonding between ligand and protein
- Hydrogen bonding interactions with implicit solvent
- Metal-chelator interactions.
All interaction potentials in Chemgauss are initially constructed using step functions to describe the interaction of atom pairs (or other chemical points) as a function of distance. These interactions are mapped onto a grid that is then convoluted with a spherical Gaussian function, which smoothes the potential making it less sensitive to small changes in the ligand position. Smoothing the score in this way serves two purposes. First docking can be run at lower resolution than would be required if the score were not smooth since small changes in position to do not cause large changes in score. Second it reduces the error associated with the rigid protein approximation by effectively accounting for the ability of the protein to make small structural re-arrangements to accommodate the ligand.
Shape interactions in Chemgauss are based on a united atom model (i.e. only heavy atoms are relevant to the shape calculation). Each ligand heavy atom is assigned a fixed clash penalty score if the distance between it and a protein heavy atom is less than the sum of the VdW radii, otherwise it is assigned a score proportional to the count of the number of protein heavy atoms within 1.25 and 2.5 times the sum of the VdW radii (atoms within 2.5 count one tenth as much as those within 1.25). From this score a penalty equal to two close protein atom contacts is subtracted to represent the VdW interactions with solvent water that are lost when the ligand docks. This score is pre-computed at grid points throughout the active site and the resulting grid is then smoothed.
Hydrogen bonding groups are modeled with one or more lone-pair or polar-hydrogen position(s) that describe the directionality of potential hydrogen bonds (with respect to the hydrogen bonding group’s heavy atom). Donor groups have lone pair positions representing the possible location of the donor hydrogen atoms relative to the donating molecule, while acceptors have lone-pair positions representing the possible locations of the donated hydrogen relative to the acceptor. A hydrogen bond is detected and assigned a constant score when a hydrogen bonding position on the ligand is within 1.0 Angstrom of a complementary hydrogen bonding position on the protein (i.e. when the polar-hydrogen position of a donor overlaps the lone-pair position of an acceptor). If the ligand hydrogen bonding group has multiple polar-hydrogens and/or lone-pair positions (groups can be both donors and acceptors) then this calculation is performed for each position and the result is summed. As with all Chemgauss terms the hydrogen bond potential is pre-computed at grid points throughout the site and then smoothed.
Hydrogen bonds to solvent molecule that break when the ligand docks into the active site are penalized by the Chemgauss scoring function. Broken protein-solvent hydrogen bonds are accounted for by calculating how many hydrogen bonds water can make with the protein at the position of each heavy atom of the docked ligand, and a penalty score is assigned which is proportional to the number of hydrogen bonds. Broken ligand-solvent hydrogen bonds are accounted for by calculating desolvation positions around each hydrogen-bonding group on the ligand that represent the positions water could occupy when making a hydrogen bonding interaction with the protein. A penalty is then assessed that is proportional to the number of desolvation positions that can no longer be occupied by water because the water in these positions would clash with the protein. As before, this potential is placed on a grid and smoothed.
Chelating interactions between protein metals and ligand chelating groups are accounted for by Chemgauss (protein-chelator and ligand-metal chelating interactions are not). For each chelator on the ligand one or more chelating-positions are calculated. If a protein metal is within 1.0 Angstrom of any chelating-position of a chelating group then a fixed score is assigned, otherwise a zero score is assigned. As before this potential is placed on a grid and smoothed.
The Chemgauss4 is a modification of the Chemgauss3 scoring function that has improved hydrogen bonding and metal chelator terms (The shape and implicit solvent interaction terms are identical to those in Chemgauss3). The new hydrogen bonding and metal chelator terms have better perception of the directionality of these interactions and also account for hydrogen bond networking effects.
To calculate the hydrogen bonding score for a ligand-protein hydrogen bond two distances are measured.
- How far the donor heavy atom is from the position the acceptor atom would consider to be an ideal for a hydrogen bonding to form.
- How far the acceptor heavy atom is from the position the donor atom would consider to be ideal for a hydrogen bonding interaction to occur.
The score for the hydrogen bond interaction is a product of two Gaussian functions of these distances scaled by the strength of the hydrogen bonding groups involved in the interaction.
HBscore = strength*g(distance1)*g(distance2)
To compute the total hydrogen bonding score for the ligand-protein complex the individual pairwise scores are calculated for all protein-ligand donor-acceptor pairs. Individual HB interaction are then eliminated if either the donor or acceptor exceeds the maximum number of interactions allowed (e.g., a hydroxyl with one hydrogen is not allowed to make more than one donor interaction), with the lowest scoring interactions eliminated first. The final hydrogen bond score is then calculated by summing the scores of the remaining individual acceptor-donor interactions.
Chemical Gaussian Overlay¶
The Chemical Gaussian Overlay function (or CGO) is primarily a ligand-based scoring function although some information from the protein structure is used as well. The similarities computed are based on the overall shape of the molecules as well as the position of hydrogen bonding and metal chelating groups. This scoring function requires a bound ligand pose along with the structure of the target protein. Typically the ligand structure is obtained from X-ray crystallography along with the structure of the target protein, although a docked ligand could also, in principal, be used.
CGO represents molecules as a set of spherical Gaussian functions describing the shape and chemistry (acceptors, donors and chelators) of the molecule. The Gaussians representing the shape of the molecule are centered at the heavy atom positions, those for donors are centered on polar-hydrogen positions (i.e. positions where the donating hydrogen could be when it is involved in a hydrogen bond), those for acceptors are centered on lone-pair positions (i.e. positions where a donating hydrogen could be when a hydrogen bond is formed) and those for chelators are centered at chelating positions (i.e. locations where a metal could have a chelating interaction). The overlap of the Gaussians on the docked ligand to those on the bound ligand are computed for each type of Gaussian (e.g. shape, donor, acceptor and chelator) by summing the overlap of individual pairs of Gaussian. The overlap of each individual pair is calculated by integrating the product of the two. To prevent chemistry not relevant to binding from contributing to the overall score, when calculating the chemistry overlaps (i.e. acceptor, donor and chelator) only groups that make the interaction with the protein are accounted for (e.g. a chelator that does not interact with a metal on the protein is ignored in the overlap calculation). The sum of all four types of overlaps is the CGO score.