PlayMolecule® AceDock: Protein-ligand docking and virtual screening [TUTORIAL]
Protein-ligand docking is, together with protein folding, one of the grand challenges in computational chemistry. Multiple software options exist that perform such task, but our favorite is rDock [1]. Its speed, accuracy, the beautiful geometry of the poses it predicts and all the different protocols that it offers (free, restrained and tethered docking, among others) makes it a great and versatile tool.
We have worked with rDock for quite some time now and we have build several functionalities around it that we would like to share with the community. These functionalities include:
- A free, web-based, graphical user interface (GUI), available at https://www.playmolecule.org/AceDock/
- Simplified tethered docking protocol. The user just needs to provide a template ligand and a series of molecules which share a common scaffold with the template. The app will take care of the alignment and the docking.
- A convenient way to add restraints by simply uploading a .csv file.
- Pharmacophoric overlap scoring. AceDock will automatically identify the main ph4 features of the template ligand you provide (aromatic rings, hydrogen bond donors and acceptors, etc.) and, after docking, the software will assign, to each predicted pose, a score based on how well the features of the pose match those on the template. This is useful for scaffold-hopping and to improve VS enrichment.
- Some files to use as examples and help you get started!
Grab a cup of coffee and walk with me trough some of these functionalities!
Free docking
This is the simplest protocol, where your ligand is allowed to explore the defined cavity in all degrees of freedom (translations, rotations and dihedral angles). To define the cavity, you must provide a template ligand (aka reference ligand) as an SDF file. This ligand should be bound to the pocket of interest. Internally, rDock will place a sphere of a radius specified by the user (Probe radius) around each atom in the template ligand, then, the space defined by this set of spheres becomes the docking cavity. If you don’t have any bound ligand for your target of interest, you can also provide a dummy atom as the template ligand, just place it on the binding site and ensure you use a bigger probe radius (>15). The detailed explanation of how the cavity is defined is provided on rDock’s manual, section 5.2.
For each ligand in the SDF file you provide in the “Ligands” input field, rDock will explore different ligand conformations, orientations and positions (always inside the boundaries of the defined cavity) and each pose will be assigned a score based on the quality of the contacts it establishes with the protein residues. Again, you can read more about the details of this process on the manual.
Let’s try it using the example files provided in the app:
Go ahead and download all the files in the Examples tab (protein, template ligand, library of ligands to dock and a .csv with some restraints). Please bear in mind that the template ligand we will use is the actual X-ray structure found in PDB code: 5UFX. One of the ligands we include in the library is that very same ligand, which allows users to easily test if they get the right pose. The second ligand is a small variant of the first, where a O atom is substituted with a NH group.
Let’s unselect the “Pharmacophoric rescoring” option for now, to keep things simple, and leave all the other options as default. In less than two minutes, you will have the results.
Hmmm… The pose ranked 1st by rDock for compound outlig1 in the library is not very similar to its actual crystal pose (the one we used as template). Docking algorithms are usually capable of sampling the right pose (defined as having an R.M.S.D. lower than 2 Å to the crystal) but they have troubles ranking it among the best sampled solutions, due to limitations in the scoring functions. In fact, if you download the results and inspect all the 10 predicted poses, it is likely that some of them will be very close to the X-ray. In my case, it was pose #6:
One way to circumvent this issue is using a pharmacophore. If you have the X-ray structure of your target bound to a small molecule, you can check if any of the poses predicted for your compounds resembles the contacts in the X-ray. If one of the predicted poses places an aromatic ring in the same spot as was found in an X-ray for a different small molecule… that’s a very good sign! Think of it in terms of the lock-key model. If your query ligand (key) is able to press the same pins of the lock (pocket) as the real key (bound molecule in X-ray structure) you might be able to open the lock! Let’s try to run this same docking exercise with the pharmacophoric rescoring.
Definitely better. In this case, we are cheating, because we are docking the same molecule we use as template, but you get the idea. Maybe try to dock a large library of molecules (~1000) against this template ligand and see what you get! Maybe you discover a new scaffold with great affinity against this target.
Scaffold docking
Somewhere along the drug discovery pipeline, you will be trying to synthesize different variants of your hit compound to get higher affinity (lead optimization). Synthesizing is expensive, so why not give a fair trial to in silico tools? If you know how your hit compound binds, you can get very accurate pose predictions for your ligands, as long as those ligands are congeneric to the template (same scaffold). Let’s see how:
Internally, our algorithm will map each atom in the query ligand to its equivalent on the template. They will be aligned, so that they occupy the same positions and, finally, rDock will do some refinement and assign a position to those atoms for which there is no equivalent on the template.
The poses look much closer to the template now. That’s the benefit of including prior knowledge on your docking runs.
Restrained docking
Some interactions are particularly challenging for docking algorithms to model. Say, for instance, water or metal coordination. One way to deal with these scenarios is to impose a restraint and ensure that a proper pharmacophoric feature in the ligand sits nearby the coordinating metal. If there is a Zn2+ ion in your pocket, you might want a negatively charged group in your ligand to interact with it. Or maybe you happen to know that the presence of an hydrogen bond donor is needed in a particular subpocket to inhibit a protein. You can do that by using restraints:
By providing this .csv file to rDock, we will force the presence of a cationic (Cat) group at position [-5, -32, 18] within a radius of 0.5 Å as well as score very favorably the presence of an aromatic group at position [0, -32, 27] within a 4 Å radius (notice the “mandatory” column).
We wait a couple of minutes and…
We can see how the cationic group (NH in 5-ring, bottom of the picture) occupies the same location in all predicted poses. You can read more about rDock restraints in its manual, at table 10.2. If one of the ligands in your library does not have any cationic group, it will be simply skipped.
Take home message
AceDock provides a series of convenient and useful protocols to address different scenarios in your drug discovery campaign. We hope this application is useful for you and, if you have any doubts or suggestions, drop us a line without any hesitation ;)
- Ruiz-Carmona, S., Alvarez-Garcia, D., Foloppe, N., Garmendia-Doval, A. B., Juhos, S., Schmidtke, P., Barril, X., Hubbard, R. E., & Morley, S. D. (2014). “rDock: A Fast, Versatile and Open Source Program for Docking Ligands to Proteins and Nucleic Acids”. PLoS Computational Biology, 10(4), e1003571. 10.1371/e1003571
