When the native ligand conformations, i. These results were comparable in quality to several widely used docking programs. Inspired by the promising results in ligand pose prediction and pose ranking using protein-based pharmacophore models, we extended the pharmacophore-based matching and scoring scheme into a docking program, named PharmDock Pharm acophore-based Dock ing. The docking program further optimizes the top ranked binding poses predicted from the pharmacophore-based scheme and re-scores the optimized binding poses with a widely used empirical scoring function.
We also present an open-source graphical user interface GUI adapted to PyMOL [ 17 , 18 ] that we have developed for PharmDock for ease use of the docking software by the scientific community. In addition, we developed two new features within the PyMOL GUI allowing the users to guide the docking process towards specific residues identified from previous experimental data. It samples the ligand binding poses by enumerating all possible multiple-points matches between pharmacophores of an ensemble of pre-generated ligand conformations and protein-based pharmacophores.
The sampled binding poses are then ranked using a simple pharmacophore-based scoring function. A set of top ranked binding poses will be locally optimized within the protein binding site to obtain the final ligand binding pose and binding score. The first two parts of PharmDock are based on a pharmacophore or functional group representation of ligand and protein whereas the last step of pose optimization is atom-based.
The details of pharmacophore generation, ligand poses sampling and ranking, final poses optimization will be described below. Overview of PharmDock. The inverse triangle shows that with progress in the overall docking process, the number of feasible binding poses of a ligand will be reduced.
[Full text] Structure-based three-dimensional pharmacophores as an alternative to | JRLCR
The overall docking process contains the following subsequent steps: Poses sampling: PharmDock samples binding poses by enumerating all possible multiple-points matches between the pre-generated ligand and protein-based pharmacophores and subsequent alignment of common features. Both Poses Sampling and Poses Ranking are based on the representation of potential interactions as potential pharmacophores. Poses optimization: The top ranked binding poses will be locally optimized within the protein binding site to generate the final prediction of ligand binding pose and binding score.
Protein-based pharmacophores refer to the potential interaction sites for the ligand to favorably interact with the protein atoms in the binding site. They can be viewed as the negative or complementary image of the topology and physicochemical properties of the protein binding site. In addition, the exclusive volume of the protein is also represented by the so-called forbidden pharmacophores, representing the binding site residues that would sterically overlap with ligand atoms placed at this respective position.
The detailed method in generating protein-based pharmacophores has been described in our previous paper [ 10 ]. A 3D grid with 0. The interaction potentials for hydrogen-bonding and hydrophobic probes placed at the grid points were computed using a continuous form of the ChemScore [ 19 , 20 ] scoring function. The aromatic and ionic interactions were calculated using a functional form similar to ChemScore. The detailed equations can be found in the previous publication [ 10 ]. The pharmacophores were generated using the computed interaction energies with the probes on the 3D grid points.
The hydrophobic pharmacophores were computed by a k-means clustering over all grid points with favorable hydrophobic scores. For each cluster, the hydrophobic pharmacophore element was then defined as the energy-weighted geometric center of all grid points of a particular cluster. The number of clusters, k, was adjusted until the minimum distance between a cluster center i and any other cluster center was on average smaller than a certain distance cutoff.
K-means clustering to generate hydrogen-bond, aromatic and ionic pharmacophores was performed over the grid points associated with the same nearest functional group.
For example, in generating a hydrogen-bond donor pharmacophore, the program iterates through all protein acceptors, and groups the grid points closest to the same acceptor into one patch. K-means clustering was then performed within this patch. In our previous study [ 10 ], we have investigated the influence of clustering distance cutoff of each pharmacophore type on the ligand pose sampling accuracy and efficiency.
We found that pharmacophore models comprised by only hydrophobic and hydrogen bond elements, which are generated using a distance cutoff of 1. These values will be used for the pharmacophore-based pose sampling process described below. The rationale is that the densest pharmacophore model provides the best description of the potential protein-ligand interactions and consequently should provide the largest amount of information for scoring.
PharmDock uses the low-energy conformers for a ligand generated by Openeye Omega [ 12 - 14 ] as docking input. Duplicate conformers are removed using a 0. The in-house program clusterconformer is then used to generate the pharmacophore elements for each ligand conformation. Hydrogen-bond pharmacophores are placed at the position of potential donor and acceptor groups of the ligand: Hydrogen-bond donors are polar hydrogen atoms bonded to oxygen, nitrogen and sulfur atoms, acceptors are oxygen, nitrogen and sulfur atoms with at least one lone pair.
The hydrophobic atoms from each ligand conformation are clustered using hierarchical clustering with a minimum distance between cluster centers of 2. Clustering is performed to reduce the number of hydrophobic ligand pharmacophores.
This significantly reduces the cost of clique detection and consequently increases the efficiency of the docking process. Aromatic pharmacophores are defined as centers of aromatic rings. Ionic groups included functional groups that are formally charged positive or negative, e. The binding pose sampling and ranking process of PharmDock has been described and discussed in our previous publication [ 10 ]. To provide the best compromise between accuracy and efficiency, only hydrophobic and hydrogen bond pharmacophore elements were used for the pose sampling process.
- Scotland: The Autobiography: 2,000 Years of Scottish History by Those Who Saw it Happen!
- Download Pharmacophores And Pharmacophore Searches Volume 32 !
- Pharmaceutical Stress Testing?
- If There Is No God: Meditations On Believing?
- Monotheism and tolerance : recovering a religion of reason.
- Taxing Wages 2009-2010!
This is in contrast to an average of four H-bond and ten hydrophobic interactions that were observed in the same dataset. A detailed evaluation of the pose sampling process can be found in our previous paper [ 10 ]. First, the length of the edge between each pair of ligand pharmacophores is determined. The edge length is also determined for each protein-based pharmacophore pair. Throughout the matching process, a tolerance of 0. The matching process can be represented by a graph in which each node represents a matching ligand-protein pharmacophore pair.
Services on Demand
The clique detection algorithm then identifies all the completely connected subgraphs from this graph. The Kabsch algorithm [ 23 ] is then used to spatially align the ligand pharmacophore elements to the matching protein-based pharmacophores in each clique, thus placing the ligand into the protein binding site. To avoid steric clashes between ligand and protein atoms, the number of heavy atoms of the ligand that are located within 1. The ligand poses sampled by PharmDock are initially scored and ranked using a simple geometric function based on the matching pharmacophore pairs formed by each ligand pharmacophore and its closest protein-based pharmacophore of the same type:.
A detailed description of the optimization procedure can be found in our previous publication [ 10 ]. The pharmacophore-based pose sampling and ranking scheme was deployed to efficiently filter out ligand binding poses that are unlikely to be the native binding pose. As shown in our previous study [ 10 ], this procedure was indeed very effective to enrich the ranking list of binding poses with native-like poses within the top positions. To further optimize the predicted ligand binding poses and estimate the binding energy of the ligand to the target, we perform a local Metropolis Monte Carlo MC [ 24 ] optimization for the top ranked poses with the Piecewise Linear Pairwise PLP scoring function [ 25 ].
PLP [ 25 ] is one of the earliest developed empirical scoring functions. It is a sum of pairwise interactions between protein and ligand atoms based on their interactions types.
- As One Is: To Free the Mind from All Conditioning!
- Numerical Approximation Methods: π ≈ 355/113?
- Parenting and Partnering with Purpose?
- A Guide to Special Education Advocacy.
Only hydrogen bonding H-bond and steric interactions are considered:. A detailed description of the PLP function can be found in Gehlhaar et al. Each pair of ligand-protein atoms is then assigned with one and only one of the interaction types: H-bond or steric. Despite the simplicity of PLP, it was shown to be one of the best scoring functions in identifying the true ligand binding poses among decoy poses and ranking the different ligands that bind to the same protein according to their binding affinities [ 26 ].
Therefore, we finally settled on PLP scoring function for our local optimization and final scoring of the ligand binding poses. An individual MC optimization was performed for every of the top binding poses generated by clique detection and ranked by the simple pharmacophore-based scoring function.
Throughout MC refinement for a given ligand binding pose, a new ligand pose is generated from the previous pose by applying small perturbations on its position, orientation and torsion angles. The newly sampled pose is accepted based on several criteria: 1 If the RMSD of the current ligand pose compared to the starting pose is larger than 3.
If Score cur is smaller than Score min , the current step is accepted. Otherwise, the Score cur is compared with the score of the last accepted step Score pre and the current pose is accepted based on the probability P :. If this factor is larger than a random number generated from a uniform distribution between 0 and 1, the current pose is accepted.
During the MC simulation, several strategies are implemented to avoid getting trapped in a local minimum or pose that has steric overlap with the protein. First, the size of maximum allowed translational, rotational and torsional changes at each MC step is adapted using the previously observed acceptance rate.
If the acceptance rate over the last MC steps is below 0. Second, if the program rejects the pose from three consecutive MC steps, the ligand is set back to its original conformation. Third, different maximum step sizes for central and terminal torsions are determined based on the equation:.inopinkear.gq
Pharmacophores and Pharmacophore Searches, Volume 32
N atoms is the total number of heavy atoms attached to the smaller branch of this torsion. The plugin features the generation of ligand libraries by exporting all ligand objects present in a PyMOL session. A lexicon of exported libraries is stored and each library of compounds can later be modified, combined with other libraries, and imported for docking to different target proteins. Illustration of PharmDock main menu with different sub-processes.
Exporting ligand libraries, performing docking calculations, either unbiased or biased, and importing docking results into PyMOL for visual analysis were shown.