Keywords:

1 Introduction

Molecular docking refers to an approach that predicts the binding mode and affinity between two interacting molecules. This approach has been widely applied to protein–ligand binding, protein–protein binding, and protein–nucleic acid binding. Molecular docking is also an important tool for structure-based drug design [1–6]. Given a potential drug target with a known three-dimensional atomic structure, a key step for drug design is to find small molecules that can bind tightly to a specific site on the target and enhance (or inhibit) the function of the target. Due to its high efficiency and low cost, molecular docking is often used for the screening of large chemical libraries for drug candidates. The top-ranked compounds from in silico screen are normally evaluated in biological assays; the confirmed active compounds are advanced for further lead optimization.

One of the examples of molecular docking tools is MDock, a protein–ligand docking suite released by our laboratory in 2007 [7]. MDock docks a rigid ligand to the protein by matching a subgroup of the ligand atomic centers to the sphere points that represent the negative image of the binding pocket, a strategy that was proposed by Kuntz and co-workers [8, 9]. Each docked pose is scored in combination with local optimization by using ITScore, an iteratively derived knowledge-based scoring function [10–12]. The pose with the lowest score is considered as the predicted binding mode, and the corresponding score is considered as the predicted binding energy score. Specifically, to account for ligand flexibility, multiple low-energy ligand conformers are pre-generated, and each conformer is docked to the protein independently. The docked conformer with the lowest score among all the conformers being docked is set as the predicted binding mode for the ligand, and the corresponding score is the predicted binding energy score.

To account for protein structural variations during ligand binding, MDock also allows users to dock a ligand simultaneously to multiple protein structures (up to 99 structures), a procedure referred to as ensemble docking. The ensemble docking algorithm in MDock is computationally efficient, with a computational time comparable to single protein docking [7, 13].

In this chapter, we will describe the methodology and the usage of MDock for molecular docking and in silico screening in detail. To further illustrate the usage of MDock for in silico screening, we will use the designed ligands for spleen tyrosine kinase (SYK), which were donated by GlaxoSmithKline (GSK) to the Community Structure-Activity Resource (CSAR) 2014 benchmark (http://www.csardock.org/) [14–16], for a case study.

2 Materials

2.1 The MDock Package

The MDock suite is freely available to academic users through application at http://zoulab.dalton.missouri.edu/mdock.htm. MDock is an open source software written in Fortran. For users who prefer to modify MDock’s source code, the Intel Fortran compiler is required for compiling the executables. For users who apply MDock directly to docking studies, the pre-compiled executable files (MDock,clu_sph, and get_sph) are provided in the bin directory of the MDock package and are ready for use. Here, MDock is the command for docking, get_sph is for selecting the sphere points that cover the binding region, and clu_sph is for clustering the sphere points from multiple protein structures for ensemble docking. The source codes are placed under the Source_ codes directory. The documentation and the demo parameter files are under the Manual directory. The tutorial files are under the Tutorial directory. Linux users can run the programs directly after MDock is installed and the bin directory is added (e.g., by typing source Install_MDock or by adding the path manually). For windows users, a Windows Linux emulator Cygwin (https://www.cygwin.com/) is recommended.

2.2 The Overview of the MDock Methodology

MDock implements a similar method to UCSF DOCK [17] to orient the rigid ligand to the binding pocket by exhaustively matching [7] the ligand atomic centers to the sphere points that represent the negative image of the binding pocket. The uniqueness of MDock lies in its iteratively derived scoring function (referred to as ITScore) and its ensemble docking method. MDock is convenient for in silico screening.

2.2.1 The Iterative Scoring Function

ITScore is a knowledge-based scoring function that was originally developed in our laboratory in 2006 [10, 11]. The main idea for the extraction of ITScore is to iteratively adjust the pairwise potentials by comparing the experimentally observed pair distribution functions (\( {g}_{ij}^{\mathrm{obs}}(r)\Big) \) derived from the native structures and the predicted pair distribution functions (\( {g}_{ij}^k(r)\Big) \) derived from the sampled decoys (including the native structures), with each decoy carrying a Boltzmann weight calculated from the interaction potentials of the current step. Finally, (\( {g}_{ij}^k(r) \)) converge to (\( {g}_{ij}^{\mathrm{obs}}(r) \)) with all the native structures having the lowest energies in comparison with their decoys. The idea is described by the following equations:

$$ \begin{array}{cc}{u}_{ij}^{k+1}(r)={u}_{ij}^k(r)+\Delta {u}_{ij}^k(r),& \Delta {u}_{ij}^k(r)=\frac{1}{2}{K}_BT[{g}_{ij}^k(r)-{g}_{ij}^{o\mathrm{bs}}(r)],\end{array} $$
(1)

where i and j represent the atom types of an atom pair, respectively, and r is the distance between the atom pair. K B is the Boltzmann constant, and T is the temperature. \( \left\{{u}_{ij}^k(r)\right\} \) are the pairwise potentials in the kth step, and \( \left\{{u}_{ij}^{k+1}(r)\right\} \) are the updated potentials for the next step. Given a set of initial potentials \( \left\{{u}_{ij}^0(r)\right\} \), the potentials are updated using the above iterative equation, until \( \left\{ {g}_{ij}^k(r)\right\} \) converge to \( \left\{{g}_{ij}^{\mathrm{obs}}(r)\right\} \) and all the native structures are associated with the lowest energies compared to their corresponding decoys. The detailed description of the iterative method is provided in [10].

It should be noted that we recently improved the scoring function by using the refined set of PDBbind 2012 [18, 19] as the new training set, which is much larger than the original training set [20]. To reproduce the results of the case study in this chapter, one should use the latest version of MDock (Ver. 2.0).

2.2.2 The Ensemble Docking

MDock implements the ensemble docking method to account for protein structural variations during ligand binding. Specifically, multiple protein structures are superimposed with the protein conformational state treated as an additional dimension for parameter optimization. The energy function for parameter optimization is defined as

$$ E=E\left(x,y,z,\phi, \theta, \psi, n\right), $$
(2)

where x, y, and z stand for the coordinates of the center of mass of the ligand, and ϕ, θ, and ψ stand for the three Euler angles, respectively. n represents the nth protein structure in the protein ensemble. MDock simultaneously optimizes the ligand coordinates and the protein conformational variable n to automatically select the optimal protein structure that best fits the ligand.

2.2.3 The In Silico Screening

MDock can be easily applied to in silico screening. For a given chemical database, the user is required to prepare a mol2 format file that provides the coordinates and Sybyl atom types of the chemical compounds. The charge and hydrogen information is not needed for MDock. The effect of charges is implicitly considered in the pairwise interaction potentials of MDock through atom types. MDock will then serially dock all the compounds onto the given target protein, predict the binding modes, and rank these compounds according to the predicted binding affinities. The top candidates can be assayed for experimental verification.

2.3 Software Dependencies

Several additional tools are also needed for file preparation for MDock. All these tools are free for academic users.

  1. 1.

    UCSF Chimera [21]:

    Chimera is used for preparing the protein and ligand files and for analyzing the docking results. The software can be downloaded directly from the website http://www.cgl.ucsf.edu/chimera

  2. 2.

    DMS (optional):

    DMS is used for building the molecular surface. DMS can be obtained from the website http://www.cgl.ucsf.edu/Overview/software.html#dms. An alternative option is to use the Write DMS tool in Chimera.

  3. 3.

    Sphgen_cpp:

    Sphgen_cpp is an accessory tool of the UCSF Dock program suite [17]. Sphgen_cpp is used for generating sphere points based on the molecular surface files. Sphgen_cpp can be downloaded from the website http://dock.compbio.ucsf.edu/Contributed_Code/sphgen_cpp.htm.

  4. 4.

    OMEGA (optional) [22, 23]

    The OMEGA is a program suite released by OpenEye Scientific Software (Santa Fe, NM, USA, http://www.eyesopen.com/). The software is free for academic users. OMEGA is used for generating multiple conformations for a given ligand. The input for OMEGA can be either three-dimensional (3D) structures in pdb format or SMILES strings in smi format. One may also use other programs to sample different ligand conformers.

3 Methods

MDock requires the 3D structure of the protein target (or an ensemble of protein structures), the 3D structure of the ligand, and the file that contains the sphere points which represent the negative image of the binding pocket. The preparation of these files is described as follows:

3.1 Preparation of the Protein and Ligand Files

The structures of the protein and the ligand can be either the experimentally determined or theoretically modeled structures. MDock uses the SYBYL mol2 format files for docking. However, for the preparation of the aforementioned sphere points, the pdb file of the protein structure is also required. The pdb file and mol2 file can be easily converted from one to the other using Chimera. Multiple structures of the ligand or multiple ligands can be stored in a single mol2 file. MDock docks the multiple structures in the ligand mol2 file one by one. For ensemble docking, the multiple structures of the protein need to be superimposed together, which can be done by the MatchMaker tool in Chimera. The protein structures for ensemble docking can be NMR models, protein structures bound with different ligands, or conformations sampled by computational techniques such as Molecular Dynamics (MD) or Monte Carlo (MC) simulations.

It is noted that solvent molecules, ions, and other co-bound small molecules should be deleted when preparing the protein structures for docking. MDock does not require the addition of hydrogens and charges for the protein and the ligand. The hydrogen and the charge information in the input files is automatically ignored.

3.2 Generation and Selection of Sphere Points for Docking

It takes the following steps to generate and select sphere points for docking purpose:

  1. 1.

    Generating the molecular surface of the protein structure.

    The molecular surface of the protein structure can be generated using the following command:

    $$ dms\ protein.pdb -a -n -o\ protein.ms $$

    where dms is taken from the molecular surface generation software DMS, protein.pdb is the pdb file of the protein structure, and protein.ms is the output file which contains the coordinates of the dots representing the molecular surface of the protein. Alternatively, Chimera’s Write DMS tool can also be used for sphere point generation.

  2. 2.

    Generating sphere points based on the molecular surface of the protein structure.

    The file contains the sphere points for the protein structure can be generated by

    $$ sphgen\_cpp - i\ protein.ms - o\ protein.sph $$

    where sphgen_cpp is the executable of the Sphgen_cpp program in the UCSF Dock software. The output protein.sph file contains the coordinates of the sphere points of the protein structure.

  3. 3.

    Defining the putative binding site.

    A pdb-format file (denoted as “site.pdb”) that locates the putative binding site is required to prepare for the selection of the sphere points that cover the binding region. For a protein structure with known binding pocket, this pdb file can be either the coordinates of the residues close to the center of the binding pocket or the coordinates of the co-crystallized ligand(s). For a protein structure with no prior knowledge of its binding pocket, users can use any binding site prediction tools or servers, such as Q-SiteFinder [24], 3DligandSite [25], and GalaxySite [26], to predict the binding pocket.

  4. 4.

    Selecting the sphere points that adequately cover the binding region.

    The sphere points which cover all the binding region can be selected using the following command:

    $$ get\_sph\ site.pdb\ protein.sph $$

    where get_sph is an accessory command in the bin directory, which selects all the sphere points within a specified distance (default: 3. 0 Å) from the atoms in “site.pdb.” The default output files are recn.sph, recn.pdb, and sph.par. recn.sph contains the selected sphere points that will be used by MDock in the docking calculations. recn.pdb is for the display of the sphere points in recn.sph using Chimera. sph.par saves the record of the parameters used by get_sph.

Users should display recn.pdb in Chimera to examine whether the binding region was adequately covered. If not, users should use a larger cutoff for sphere points selection. The cutoff distance for sphere points selection and other parameters for get_sph can be specified in two ways:

  1. 1.

    Run get_sph interactively:

    $$ get\_sph\ site.pdb\ protein.sph- param $$

    Users will be asked to provide a value for each parameter. If users decide to use the default value for a parameter, simply hit “Enter.” The parameters will be output in the sph.par file.

  2. 2.

    Run get_sph with parameters defined in a parameter file, say, sph.par:

    $$ get\_\ sph\ site.pdb\ protein.sph- param\ sph.par $$

    All the parameters will then be read from this parameter file.

    A detailed explanation of the parameters in get_sph is provided in the manual of MDock.

For molecular docking against a single protein structure, the sphere points in recn.sph will be directly used for molecular docking. The preparation is more complicated for molecular docking against multiple protein structures. Specifically, the sphere points should be prepared for each protein structure independently. Then, these sphere points generated from individual protein structures are combined and clustered as follows:

$$ \begin{array}{c} cat\ */ recn.sph> all.sph\\ {} clu\_sph\ all.sph\ recn.sph\\ {}\end{array} $$

The output file, “recn.sph”, comprises the coordinates of the sphere points to be used for ensemble docking.

3.3 Molecular Docking

3.3.1 Single (Protein) Docking

The docking command, MDock, can be executed in three ways:

  1. 1.

    Run MDock using default parameters:

    $$ M Dock\ protein.mol2\ ligand(s).mol2 $$

    The ligand(s).mol2 file contains a single ligand conformer or multiple conformers of a ligand or multiple ligands. MDock automatically docks all the conformers to the protein. This method requires the sphere point file recn.sph as a standard input. The default parameters will be used for the docking calculation, the values of the parameters will be output in a parameter file named MDock.par.

  2. 2.

    Run MDock interactively:

    $$ M Dock\ protein.mol2\ ligand(s).mol2 - param $$

    MDock will interactively ask users to provide a value for each parameter. If users prefer the default value, hit “Enter” key. The input values of the parameters will be saved in a parameter file named MDock.par.

  3. 3.

    Run MDock by using the parameters pre-defined in a parameter file, say, MDock.par:

    $$ M Dockprotein.mol2\ ligand(s).mol2 - param\ M Dock.par $$

    MDock will search in MDock.par for the required docking parameters. If any required parameter is missing, MDock will interactively ask the user to specify the value for the parameter.

Besides its application to molecular docking, MDock can also be applied to binding mode optimization, scoring, and even target selectivity. For detailed descriptions of the parameters for MDock, the user can refer to the MDock manual.

3.3.2 Ensemble Docking

For ensemble docking, the mol2 files for the multiple protein structures should be under the same directory, with their file names sharing a user-defined prefix followed by a double-digit number (from 01 to 99) to label the protein structures. For example, if we have eight protein structures for ensemble docking, we define the prefix as PKA, then the eight mol2 files will be named PKA01.mol2, PKA02.mol2, …, and PKA08.mol2. The command to run MDock against multiple protein structures is

$$ M Dock\ P K A\ ligand(s).mol2 $$

Similar to single docking, ensemble docking also requires the sphere point file, “recn.sph”, and can be run interactively with user-defined parameters or with the parameters defined in a parameter file.

3.3.3 The Output Files of MDock

MDock creates three files: a mol2 file that lists the docked modes including their coordinates and energy scores (default: MDock.mol2), an output file that lists the energy scores of the docked modes (for ensemble docking, the corresponding protein structure number for each docked mode is also included) (default: MDock.out), and a file that records the information about the consumed CPU time and the number of processed ligand conformers (default: MDock.log).

4 The Case Study

The target SYK with its 276 ligands in the CSAR 2014 benchmark is used for the case study. The CSAR benchmark provides the SMILES strings of the ligands in an smi-format file (SYK_set.smi). The pIC 50 values of all the ligands for SYK and the complex structures (in pdb format) for eight ligands (GTC000222 to GTC000226, GTC000233, GTC000249 and GTC000250) are also released in the benchmark.

In our docking study, up to 500 conformers for each ligand were generated from its SMILES string using Omega 2.4.6 using the following command:

omega2 -in SYK_set.smi -out ligs.mol2 -warts true -fromCT true -strictfrags true -maxconfs 500 -flipper true

A total of 108,981 3D conformers for 276 ligands were generated and stored in mol2 files (ligs.mol2). Both single docking and ensemble docking were performed. Specifically, the protein structure from the SYK-GTC000222 complex was used for single docking. As for ensemble docking, the protein structures from the eight released complexes were superimposed with Chimera, using the protein from the SYK-GTC000222 complex as the reference structure; these eight structures formed the protein conformational ensemble.

The files containing the sphere points for single docking and ensemble docking were prepared as described in Sect. 3.2. For sphere point selection, we used the ligand coordinates (site.pdb) from the PDB entry 1XBC [27], which contains SYK bound with a small molecule that binds to the same binding pocket as those 276 ligands. 1XBC was also superimposed to the SYK-GTC000222 complex using Chimera.

Figure 1a, b show typical views of the protein structures and the sphere points being selected for single docking and ensemble docking, respectively. The parameter files used for single docking and ensemble docking are identical and are shown in Fig. 2. For each ligand conformer, up to 1000 poses were rigidly sampled followed by local optimization and scoring. Only the best scored pose for each conformer was saved. The computations were performed using a single 3.40 GHz Intel Core i7 CPU. Single docking took 56,265 s, whereas ensemble docking took 27,103 s. Ensemble docking is more efficient in this case study because by using multiple protein structures, the local optimization was more easily to converge, and because with more protein conformers, ensemble docking requires fewer than 1000 ligand poses to exhaustively sample the possible binding poses.

Fig. 1
figure 1

(a) The protein structure with the sphere points for single (protein) docking. (b) The protein structures with the sphere points for ensemble docking

Full size image
Fig. 2
figure 2

The parameter file used for single (protein) docking and ensemble docking

Full size image

For each ligand, the best-scored binding mode (i.e., the mode with the lowest score) among all the docked conformers was considered as the predicted binding mode, and the corresponding score was considered as the predicted binding energy score. The root-mean-square-deviation (RMSD) of the heavy atoms between the predicted binding mode and the native binding mode was used as the metric for evaluating the performance of binding mode prediction, whereas the Pearson correlation coefficient (r) between the predicted binding energy scores and the − pIC50 values was used to evaluate the performance of the binding affinity prediction. Table 1 lists the RMSDs of the best predicted binding mode (with the lowest RMSD) for the top 1 prediction and for the top 3 predictions, respectively, for the eight ligands with released complex structures. The results of single docking and ensemble docking are also shown, respectively.

Table 1 Results of binding mode prediction on SYK with single (protein) docking and ensemble docking
Full size table

It can be seen from Table 1 that single docking and ensemble docking show comparable performances. An example of successful ensemble docking for binding mode prediction is shown in Fig. 3: The top prediction for the ligand GTC000222 achieved an RMSD of 0. 866 Å compared with the native binding mode. For binding affinity prediction, ensemble docking shows a better performance than single docking: In the score versus − pIC50 plot shown in Fig. 4, the predicted binding affinities for the 276 ligands with ensemble docking achieved a higher correlation with the − pIC50 values (0. 72) than the correlation achieved from single docking (0. 51).

Fig. 3
figure 3

The binding mode of the ligand GTC000222 was successfully predicted by ensemble docking, with an RMSD of 0.866 Å from the native binding mode. The protein is represented by its molecular surface. The native ligand is in cyan, and the predicted binding mode is in magenta

Full size image
Fig. 4
figure 4

(a) The score versus − pIC50 plot for single (protein) docking. (b) The score versus − pIC50 plot for ensemble docking

Full size image

Other applications of MDock can be found in our publications [13, 20, 28, 29].

5 Notes

  1. 1.

    For the preparation of the protein structures for docking, solvent molecules, ions and co-bound small molecules should be removed.

  2. 2.

    MDock does not need to add hydrogens and charges for the protein and the ligand. The information about hydrogens and charges in the input protein and ligand files is automatically ignored.

  3. 3.

    For ensemble docking, the multiple protein structures should be superimposed for preparing the sphere points and docking calculation.

  4. 4.

    For sphere point selection, users should manually examine the sphere points to make sure that the whole binding region is adequately covered by the sphere points. Otherwise, the value of the cutoff distance should be increased to include more sphere points.