Introduction:

Virtual screening has become an indispensable tool in drug discovery, enabling the rapid identification of potential drug candidates through computational methods. However, the reliability and accuracy of virtual screening results often need to be validated by experimental bioassays. In this tutorial, we will explore how to compare virtual screening results obtained using PyRx with experimental bioassays. PyRx, a user-friendly virtual screening tool, allows researchers to perform molecular docking experiments to predict the binding affinity of small molecules to biological targets. By importing sample docking results and generating input files for comparison with bioassay data, you will gain hands-on experience with both virtual and experimental approaches. Ultimately, this tutorial aims to provide insights into the advantages and limitations of virtual screening and how it can complement traditional bioassay techniques in drug discovery.

Comparing Virtual Screening with Bioassays Using PyRx

In this tutorial we’ll use PyRx to compare virtual screening results with Bioassays. We’ll start by importing a sample precomputed docking result to see what to expect from a single docking experiment. Then we’ll create input pdbqt files that will be used to compare virtual screening results with bioassays. At the end of this tutorial, you’ll have hands-on experience in working with bioassay data and you’ll be able to run virtual screening experiments using PyRx. We’ll conclude with a discussion about publicly available bioassay data and current limitations and advantages of virtual screening tools.

Prerequisite: PyRx – http://pyrx.scripps.edu/downloads

Since PyRx is installed on all iMacs in this training room, we’ll start by running it from the /Applications folder or from the Dock. Users can also install PyRx on their personal laptops by going to the download page for PyRx, at the above URL.

Part 1: AutoDock Wizard

In this part we’ll do a dry run of the AutoDock Wizard, using precomputed docking results.

Exercise 1: Importing Sample Data

Use the File -> Import menu, select Workspace Tarball – Remote File, click Next, then click Finish. Wait for the Import Completed Successfully dialog and press OK.

PyRx extracts sample data and displays the directory structure for the Ligands and the Macromolecules folders using the AutoDock tab under the Navigator panel shown on the right.

Click on the arrow next to hsg1 to expand or contract the folders. As you can see from this example, PyRx creates a folder for each target within the Macromolecules folder, where it stores the docking log (dlg) files. Double-click on a pdbqt file in this widget to show the corresponding molecule in a 3D Scene.

Control+click on any of the files in the AutoDock tab and you’ll see a menu of options such as:

Edit Opens file in Documents editor Delete deletes files or directories Properties shows the path on this hard drive Refresh Refreshes directory tree

There are additional options called Display and Display (Mayavi) for pdbqt and for map files, respectively, which we will cover later on.

Before we work with the AutoDock Wizard, let’s learn how to transform molecules in the 3D Scene.

Mouse and Key Bindings

In this training room, Mac mouse preferences are set to emulate a 3-button mouse. This means that clicking on the top-right part (i.e., the Secondary Button) emulates the right- mouse click. Also, Button 3 act as a scroll wheel that can be used to zoom in and out in a 3D Scene.

PyRx uses Mayavi for the 3D Scenes, and it uses the same mechanism to manipulate 3D objects in the scene. The following is taken from the Mayavi v3.3.2 documentation to describe interactions with the Mouse:

The view in the scene can be changed by using various mouse actions. Usually these changes are accomplished by holding down a mouse button while dragging the mouse:

• holding the left mouse button down and dragging will rotate the camera/actor in the direction moved.
  • Holding down “SHIFT” when doing this will pan the scene – just like the middle button.
  • Holding down “CONTROL” will rotate around the camera’s axis (roll).
  • Holding down “SHIFT” and “CONTROL” and dragging up will zoom in, while dragging down will zoom out. This is similar to pressing the right mouse button.
• holding the right mouse button down and dragging upwards will zoom in (or increase the actor’s scale), and dragging downwards will zoom out (or reduce scale).
• holding the middle mouse button down and dragging will pan the scene or translate the object.
• Rotating the mouse wheel upwards will zoom in and downwards will zoom out.

Users can also press the ‘r’ key on the keyboard to reset the camera’s focal point and position. Read the Mayavi User Guide for a complete list of features: http://code.enthought.com/projects/mayavi/docs/development/html/mayavi/application.h tml#interaction-with-the-scene

Exercise 2: Using the AutoDock Wizard

Click on the AutoDock Wizard tab under the Controls panel to see the widget shown above. There are 3 different execution modes that PyRx currently supports: (1) Local is disabled by default. It is activated when PyRx finds autodock binaries in your “path”.

You can also use Edit -> Preferences… to tell PyRx where to find the autodock and autogrid binaries. (2) Cluster execution mode is also disabled, since we don’t have the qsub command in our path, which can submit batch jobs to a cluster. (3) Remote mode is always enabled, and it uses web services provided by the National Biomedical Computation Resource (http://ws.nbcr.net) to run remote AutoDock jobs.

Note that starting with PyRx 0.6 there is a new Vina Wizard, which is selected by default. Since Vina binaries are automatically distributed with PyRx, the Local mode for Vina is enabled by default. There is no Remote mode available for Vina in PyRx, but Vina web services might become available through ws.nbcr.net in the future.

Exercise 3: Select Molecules

Follow the instructions on the screen. Select ind.pdbqt from the Ligands folder, and select the hsg1 folder or the hsg1.pdbqt file from the Macromolecules folder. Click Forward to go to the next exercise.

Exercise 4: Run AutoGrid

In this step you’ll see a grid box in the 3D Scene and a Run AutoGrid page under the AutoDock Wizard. You can move the spherical handles of the grid box to adjust the dimensions for the AutoGrid maps. Click on the Maximize button if you want to enclose the entire molecule in the gridbox (i.e., if you do not know the location of the binding site you want to target). Click on Reset to restore the grid dimensions back to 40x40x40 and to reset the box’s center to the geometric center of the molecule.

Click Forward to continue.

Note: since we already have precomputed grid maps for this example, PyRx won’t run autogrid again. However, if we didn’t have the grid maps, or if the dimensions of the grid maps had been changed, then the Forward button would start an autogrid run. You can click on the Run AutoGrid button if you would like to run autogrid in this step, regardless of the available grid maps.

Exercise 5: Run AutoDock

The Run AutoDock tab allows you to select among four docking algorithms available in AutoDock. Click on Docking Parameters… to see the run parameters that can be changed. Users who are familiar with AutoDock can adjust these parameters to their liking. In particular, the maximum number of energy evaluations is set to 250,000 by default, which results in a short run. Users can choose either the short, medium, long or custom value for this parameter. Click on OK to close this window, then click the Forward button to go to the “analyze results” section. Similar to the Run AutoGrid page, PyRx opens precomputed dlg files, when they are available. Note: the last sentence of the status bar reads Click Forward to Analyze Results. If we did not have any precomputed dlg files for this protein-ligand complex, then this line would have read Click Forward to Run AutoDock, instead. Since we already imported the precomputed dlg file during Exercise 1, we can now go to the next exercise, where we’ll see how docking results can be analyzed.

Exercise 6: Analyze Results

The Analyze Results page is where the final docking results are presented. For the AutoDock Wizard, this page contains a table, as shown above. Users can sort this table according to the values in any particular column. Click on a row to see the corresponding docking pose in the 3D Scene. You can also export the numerical results as a Comma-Separated Values (CSV) file, which can later be imported into a third-party spreadsheet application.

Before we start the next part, let’s remove all the files from the workspace. Click on the AutoDock tab in the Navigator window, and use Shift + the mouse to select all the files in the Ligands folder. Right-click, select Delete, and click Yes to confirm.

This concludes the case study of docking a single ligand to a single receptor with AutoDock. In the next part of the tutorial, we’ll use AutoDock Vina and data from PubChem’s BioAssays to do a virtual screen and to compare the computational results with the wet-lab experiment.

Part 2: Virtual Screening and BioAssay

To compare our in silico screen with a real in vitro experiment, we’ll use

15-Hydroxyprostaglandin Dehydrogenase (PDB ID: 2GDZ) as our target. This is one of the few proteins that has both a 3D structure as well as bioassay data available in PubChem. We’ll start by preparing the input files for our virtual screen.

Exercise 1: Prepare the Input Structures

Click on File -> Import, click Next, enter 2gdz for the PDB ID, and click Finish. Right-Click on 2gdz in the Molecules tab, and select

AutoDock -> Make Macromolecule. This opens up a dialog box that can be used to select amongst alternate conformations. Click OK to let PyRx (1) create a 2gdz folder under Macromolecules and (2) put 2gdz.pdbqt in it.

Click on the Open Babel tab under the Controls pane. Click on the first icon on the toolbar, and open Desktop/PyRx2010/3D.sdf. You can also download this file from the web by following the links to this PubChem BioAssay from the PyRx blog: http://pyrx.scripps.edu/blog/81-thermal-shift-assay-for-inhibitors-of-hpgd

Right-Click on any of the entries inside the Open Babel widget, and select Convert All to AutoDock Ligand (pdbqt). This opens a progress dialog box and shows the pdbqt files created in the Ligands folder. Now that we have the input pdbqt structures, we are ready to use the Vina Wizard.

Exercise 2: Using the Vina Wizard

Click on the Vina Wizard tab under the Controls pane. The first page for the Vina Wizard is similar to the AutoDock Wizard, except that there are 2 execution modes, instead of 3 (i.e., there are currently no remote web services for Vina). Also, since Vina binaries are distributed within PyRx, the Local execution mode is enabled by default.

Click on the Start button to begin. On the Select Molecules page, use

Shift + the mouse to select all the ligands, select 2gdz under the Macromolecules folder, and click Forward. On the Run Vina page, click the Maximize button to make the Vina search space large enough to include all the atoms from our target.

Click Forward to start the virtual screen. PyRx now loops through all the selected ligands and runs Vina for each of them. You’ll see a new tab in the View pane where PyRx displays: the command line used to start Vina, the working directory, and the output from Vina. Docking each ligand takes around a minute; thus, the entire virtual screen might take an hour to finish. You can click on Analyze Results to see the Vina results as they are created. If you have access to a computer cluster with the Portable Batch System (PBS) or Sun Grid Engine (SGE) installed, then you’ll have the qsub command in your path, which will allow you to run virtual screens on a cluster. In this case (when qsub is in your path), PyRx enables the Cluster execution mode on the start page. You can also try the AutoDock Wizard with the Remote execution mode later.

Exercise 3: Comparing Docking Results with BioAssay

PyRx stores virtual screening results in a multi-table database that you can access by clicking on the Tables tab under the View pane. This table widget provides a bird’s-eye view of your workspace, and it includes tables for Ligands, Targets, and Docking Results. In addition, it has an option to import other tables that are in the Comma- Separated Values (CSV) format, such as bioassay data (or data from your own assays).

This SQLite3 database is stored in ~/.mgltools/PyRx/db.sqlite3. To save some space, only the lowest binding energies per job are stored in the Docking Results table. PyRx provides an option to save these tables as CSV files ( icon on the toolbar).

Click the icon on the toolbar to open the Data Plotting Dialog. By default, this widget opens with the Docking Results table selected, and it plots Binding Energy versus Ligand ID. If PyRx can’t convert all entries in a given column into integer or floating point numbers, then it will use the index (row) of the entries, instead. Click the drop-down menu under Select

Table, and select Ligands. Then select Torsional DOF for the Y Column and Size for the X Column. This creates the figure shown on the right. Click OK to store this figure under the 2D Plots tab. You can then zoom, pan, and save this figure using matplotlib icons on the toolbar.

Click the icon on the Tables tab, and open Desktop/PyRx2010/BioAssay.cvs. You can also download this file from the web by following the links to this PubChem BioAssay from the PyRx blog:

http://pyrx.scripps.edu/blog/81-thermal-shift-assay-for-inhibitors-of-hpgd

Click the icon again, and select Docking Results under Select Table. You’ll see a plot similar to the one shown above. Since we now have a fourth table that contains an Outcome column with “Active”, “Inconclusive,” or “Inactive” entries, PyRx checks to see if the compound IDs (CID) in this table match the CIDs under the Docking Results table’s Ligand column. If so, PyRx plots Active entries in red and Inactive entries in blue. In addition, PyRx plots a Receiver Operating Characteristic (ROC) curve, where you can see the True Positive Rate (Sensitivity) plotted against the False Positive Rate (1 – Specificity). ROC curves are very useful when comparing virtual screening results with in vitro experiments (especially when you are developing docking algorithms or testing different parameters). For instance, if we had to pick randomly from our ligand library, we would get 0.5 for the area under the ROC curve (AUC) – i.e., the diagonal dotted line. In this particular case (this Thermal Shift Assay for Inhibitors of 15- Hydroxyprostaglandin Dehydrogenase), we get perfect results with AutoDock: binding energies for all active compounds are more favorable (lower) than the binding energies for the two inactive compounds. This gives us an AUC of 1, as displayed in the above figure.

Read the PyRx blog for other examples of ROC curves: http://pyrx.scripps.edu/blog

Conclusion:

In this tutorial, we have explored the process of comparing virtual screening results from PyRx with experimental bioassays. Through hands-on practice, you have learned how to interpret docking results, generate input files, and analyze bioassay data for comparison. Virtual screening offers a fast and cost-effective approach to identifying potential drug candidates, but it is essential to understand its limitations, including the need for validation with experimental data. Publicly available bioassay datasets provide a valuable resource for such comparisons, allowing researchers to enhance the accuracy of their predictions. As we continue to refine virtual screening tools like PyRx, integrating them with bioassays will remain a critical step in the drug discovery pipeline.