Chapter 19

Small-Molecule Library Screening by Docking with PyRx

Sargis Dallakyan and Arthur J. Olson

Abstract
Virtual molecular screening is used to dock small-molecule libraries to a macromolecule in order to find
lead compounds with desired biological function. This in silico method is well known for its application in
computer-aided drug design. This chapter describes how to perform small-molecule virtual screening by
docking with PyRx, which is open-source software with an intuitive user interface that runs on all major
operating systems (Linux, Windows, and Mac OS). Specific steps for using PyRx, as well as considerations
for data preparation, docking, and data analysis, are also described.

Key words Virtual molecular screening, Computer-aided drug design, Molecular docking, PubChem,
AutoDock, Vina, Open Babel

1 Introduction

Drug discovery is an attractive research area that enables application

of cutting-edge biomedical research to improve health of many
people [1]. In the past, medicines were derived from natural prod-
ucts, mostly from plant sources. While natural products continue
to be used and researched for medicine, it is now possible to syn-
thesize a large number of chemical compounds that are not readily
available in nature. The increased number of possible chemical
compounds presents both a challenge and opportunity for the
pharmaceutical industry. Testing different drug candidates in
human clinical trials is a long and expensive process, which is why
phenotypic or target-based screening is so important in the earlier
stages of drug discovery [2].
In phenotypic screening, different compounds are tested in
cells or organisms to see which compound makes intended changes
in the phenotype. When molecular causes of the disease are
unknown, phenotypic screening is, in many cases, the only avail-
able option for finding life-saving drugs. For diseases that are well
studied and understood at the molecular level, altering a single
macromolecule can lead to a desired outcome. An example of such

Jonathan E. Hempel et al. (eds.), Chemical Biology: Methods and Protocols, Methods in Molecular Biology,
vol. 1263, DOI 10.1007/978-1-4939-2269-7_19, © Springer Science+Business Media New York 2015

243
244 Sargis Dallakyan and Arthur J. Olson

a macromolecular target for the common flu virus is discussed

shortly. In target-based screening, compounds are tested with
purified macromolecules (usually a protein) to find lead com-
pounds that make intended macromolecular changes. For a lead
compound to become a drug, it needs to be able to reach a site of
action in the body, bind to its target macromolecule, and elicit the
desired biological effect.
Compared to large biological molecule therapeutics, such as
insulin or antibodies, which are administered through injection,
small molecules can be taken orally and are better at reaching
different sites in the body. This is why the majority of approved and
experimental drugs are small molecules. Small molecules are also
better suited for virtual molecular screening, which is the main
subject of this chapter. With virtual screening, different compounds
are docked from a small-molecule library to a target macromole-
cule (usually a protein) to find compounds with the best binding
affinity [3]. Note that virtual screening is not limited to drug tar-
gets and it can be used to screen against herbicides, pesticides, or
any other target of interest [4]. In all cases, finding the right target
is very important for virtual screening campaign to succeed. When
the three-dimensional (3D) structure of a target is available,
through X-ray crystallography, NMR spectroscopy, or any other
means, docking algorithms can be applied to search for the best
binding mode between target macromolecule and ligand.
In this chapter, methods for performing virtual screening
experiments with PyRx open-source software are outlined. The
3D structure of the influenza virus neuraminidase [5] is used as
an example to show how to prepare an input file for the target
macromolecule. Influenza virus neuraminidase cleaves sialic acid
from the infected cell surface to release newly created viruses.
Neuraminidase inhibitors bind to neuraminidase and prevent
them from binding to sialic acid. This leaves the influenza virus
stuck on the surface of infected cells, so that the influenza virus
cannot infect nearby healthy cells [6]. Here, steps to prepare
input structures for zanamivir (a neuraminidase inhibitor), sialic
acid, and sucrose (table sugar) are described. These small mole-
cules are then used to run virtual screening against influenza virus
neuraminidase.

2 Materials

PyRx is written in Python programming language and it can run

on nearly any modern computer, from PC (personal computer) to
supercomputer. Below details are provided of the Windows PC
used in Subheading 3, although similar methods also work on
Linux and Mac OS as well.
 Small-Molecule Library Screening by Docking with PyRx 245

2.1 Hardware 1. Dell Studio 540S with Intel Core 2 Duo CPU at 2.53 GHz,
and Software 4 GB memory (RAM), ATI Radeon HD 3400 series graphics
card, and 32-bit Windows Vista operating system.
2. Binary distribution of PyRx version 0.8 for Windows available
free from http://pyrx.sourceforge.net.

2.2 Input Files To start with structure-based virtual screening, structures of the
target macromolecule and small molecules are needed as input
files. There are a number of publicly available websites where users
can download these input files. Used in this chapter are DrugBank
[7] to get the structure of zanamivir, PubChem [8] for 3D structure
of sucrose, and Protein Data Bank [9] to get 3D structures of
influenza virus neuraminidase and sialic acid.
1. Open a preferred web browser and visit http://www.drug-
bank.ca/drugs/DB00558, click on SDF link next to
Download, and save that page as DB00558.sdf.
2. Go to http://pubchem.ncbi.nlm.nih.gov/summary/summary.
cgi?cid=5988, click SDF icon on top right, and select 3D SDF: Save.
3. Visit http://www.pdb.org/pdb/explore/explore.do?structure
Id=2BAT, click on Download Files, and select PDB File (Text).
The reason for choosing these particular molecules is that they
are familiar to most of the readers and computations can be run
relatively quickly on a PC. To apply the protocol described in
Subheading 3 to other binding target and ligands, users would
need to obtain input files corresponding to their specific binding
target and ligands. Selection of the binding target depends on the
biological problem of interest, and it is assumed that the 3D struc-
ture of the target is available in PDB format through Protein Data
Bank (http://pdb.org) or other sources (see Note 1). Selection of
ligands depends on whether virtual screening is used for lead dis-
covery or lead optimization. For lead discovery, it is advised to
include as many ligands with diverse shapes, sizes, and composition
as possible. Since individual docking computations are indepen-
dent from each other, users are practically only limited by compu-
tational power available at their disposal. For lead optimization, on
the other hand, ligands are selected to closely match the lead com-
pound [10]. One of the advantages of virtual screening is that it is
not limited to commercially available compounds; a ligand file for
a novel compound not found in any of the databases can also be
used (see Note 2).

3 Methods

3.1 Prepare Input Before input files can be used for virtual screening, they must be
Files for Docking converted to the PDBQT file format suitable for docking with
AutoDock Vina [11].
246 Sargis Dallakyan and Arthur J. Olson

1. Start by double-clicking on PyRx icon on the Desktop.

2. Select Open Babel tab under Controls panel and click on the
first icon on its toolbar with plus (+) sign on it. Navigate to the
Downloads folder and select CID_5988.sdf (sucrose from
PubChem).
3. Click on the first icon on the Open Babel toolbar again, and
locate and open DB00558.sdf (zanamivir from DrugBank).
558 is the accession number of zanamivir in DrugBank and it
is listed under the Title column in the Open Babel table. If
other molecules are to be included in virtual screening, the
Open Babel widget can be used to convert them to PDBQT
file format (see Note 3).
4. Select the row corresponding to zanamivir with Title 558, and
right-click and use the Minimize Selected option. Click OK
and wait for energy minimization to complete. Notice that the
title of this molecule has changed to 558_uff_E = 197.68. The
_uff part corresponds to the force field used for energy mini-
mization, which, by default, is the Universal Force Field [12]
as implemented in Open Babel software package [13]. The
_E = 197.68 part corresponds to the energy of the minimized
molecule. The precise value for this energy is not important
here. However, this notation is helpful to capture changes
made to this molecule before conversion to the AutoDock
ligand file in the next step.
5. Right-click on any of the rows in Open Babel table and use
Convert All to AutoDock Ligand (pdbqt). This will create two
pdbqt files corresponding to sucrose and zanamivir molecules
under the Ligands folder.
6. Select Documents tab under the View panel, click on the Open
icon (second from the left), and open the 2BAT.pdb file. 2BAT
is the PDB ID for the structure of the complex between influ-
enza virus neuraminidase and sialic acid [5]. The following
steps are specific to this structure. To apply this method to
targets which have no ligand attached, please go directly to
step 10 and replace 2BAT with the name of the desired target
macromolecule.
7. Next select lines corresponding to sialic acid from 2BAT.pdb.
Scroll down, use Ctrl-F or the Find icon on the toolbar
to search for SIA residues, and select lines with HETATM
3216–3236. Use Ctrl-C or right-click Copy, click on the New
icon, and paste these lines (Ctrl-V or right-click Paste) in a new
file. Save this file as SialicAcid.pdb using Save icon (third from
the left) on the Documents panel. If working with another
target that contains a ligand that is desired for re-docking, the
 Small-Molecule Library Screening by Docking with PyRx 247

Documents panel in PyRx can be used or any other text

editor (such as Notepad or WordPad) to extract HETATM
records corresponding to the ligand of interest. The web page
for 2BAT (http://www.rcsb.org/pdb/explore.do?structure
Id=2BAT) also lists different ligands bound to neuraminidase,
including sialic acid, which is listed under Ligand Identifier
column as SIA. This web page also offers the possibility to
download ligand SDF file for sialic acid.
8. Click on 2BAT.pdb tab under the Documents panel, scroll up,
and left-click at the beginning of the line starting with TER
3023. The TER record indicates the end of a list of ATOM
records for a chain according to PDB file format specification.
In this case, it is desired to keep neuraminidase atom records
only and delete all other records that correspond to different
ligand and water molecules cocrystallized with this structure of
neuraminidase. With the left mouse button pressed, scroll
down to the end of the file and click Delete. Save this modified
2BAT.pdb file using the Save icon again.
9. From the menu bar, use File → Load Molecule menu and open
SialicAcid.pdb. Right-click on SialicAcid under Molecules
panel and select AutoDock → Make Ligand.
10. Use File → Load Molecule menu again and open 2BAT.pdb.
Right-click on 2BAT under Molecules panel and select
AutoDock → Make Macromolecule.

3.2 Run Virtual 1. Select Vina Wizard tab under the Controls panel and click on
Screening Using Vina the Start button.
Wizard 2. Select 558_uff_E = 197.68.pdbqt, 5988.pdbqt and SialicAcid.
pdbqt under the Ligands folder (use the Shift key for selecting
multiple ligands).
3. Select 2BAT under the Macromolecules folder and click on the
Forward button on Vina Wizard.
4. Click on the Maximize button under Vina Search Space and
then click on the Forward button. This starts AutoDock Vina
and docks each ligand, one by one, to neuraminidase (2BAT).
It takes less than 20 min to complete this virtual screening on
a PC mentioned in Subheading 2.1 (see Note 4) (Fig. 1).
5. After virtual screening is completed, PyRx automatically
advances to Analyze Results page, where results of virtual
screening computation can be viewed. AutoDock Vina, by
default, outputs the ten best binding modes for each docking
run (see Note 5). Left-click on Binding Affinity (kcal/mol)
table header cell under Analyze Results tab to sort this table by
predicted binding affinity (see Note 6).
248 Sargis Dallakyan and Arthur J. Olson

Fig. 1 A screenshot of the PyRx virtual screening tool. The table on the left lists predicted binding affinity of
zanamivir (2BAT_558_uff_E = 197.68), sialic acid (2BAT_SalicAcid), and sucrose (2BAT_5988) for influenza
virus neuraminidase (2BAT). The 3D scene on the right shows line drawing and transparent molecular surface
of neuraminidase. Ball-and-stick models for zanamivir and sialic acid are also shown on this 3D scene

4 Notes

1. During docking runs, the 3D structure of the target is fixed

while the ligand is moved and rotated to find the best binding
modes. While it is possible to make some of the side chains
flexible during the docking, incorporating full flexibility of the
target is still a subject of active research [14].
2. There are a variety of desktop or Web-based molecular editors
available that can be used to generate a ligand file for a novel
compound not found in any of the databases. The Web-based
molecular editors allow users to sketch molecules in 2D, while
desktop tools such as Avogadro [15] can draw molecules in 3D.
3. SDF (Structure-Data File) format is commonly used to store
multiple structures in a single file. It allows storing arbitrary
data together with coordinates and atom types. Oftentimes,
small molecules stored in SDF are flat (2D) and energy mini-
mization is performed to get 3D structures with proper bond
length between different atoms.
4. The main results from virtual screening runs are the best
predicted binding modes and corresponding binding affinity.
The negative values for binding affinity (or binding free energy)
indicate that the ligand is predicted to bind to a target mac-
romolecule. The more negative the numerical values for the
 Small-Molecule Library Screening by Docking with PyRx 249

binding affinity, the better is the predicted binding between a

ligand and a macromolecule. In this particular case of screening
neuraminidase with zanamivir, sialic acid, and sucrose, Fig. 1
shows that zanamivir (2BAT_558_uff_E = 197.68) and sialic
acid (2BAT_SalicAcid) are both predicted to have the best
binding affinity of −7.3 kcal/mol, whereas the best binding
mode for sucrose (2BAT_5988) is predicted to have binding
affinity of −7.0 kcal/mol. In other words, zanamivir and sialic
acid are predicted to have better binding affinity to neuramini-
dase than sucrose. The fact that both zanamivir and sialic acid
have the same predicted binding affinity indicates that zanami-
vir can bind to neuraminidase and inhibit it from binding to
sialic acid.
5. PyRx users can also export virtual screening results as CSV
(Comma-Separated Values) or SDF files. This is useful for
further analysis, filtering, or re-ranking of virtual screening
results with third-party packages.
6. There are a number of approximations used to model protein-
ligand interactions [16] and there are a number of unknowns
when it comes to comparing virtual screening results with
experiments [17], not the least of which is that a single protein
is being docked with a single ligand. In practice, even with
purified samples, it is hard to predict if proteins or small-mole-
cule ligands would aggregate and whether idealistic prediction
of binding affinity with single protein-ligand docking applies
to diluted samples. Nevertheless, small-molecule virtual screen-
ing by docking is a very valuable in silico method that can rank
small molecules according to their predicted binding affinity to
a target macromolecule. The cost of running virtual screening
experiments is minuscule compared to real screening experi-
ments. Virtual screening is also a very good tool for hypothesis
generation with which to test modified versions of existing
compounds or custom compounds that are not commercially
available. With advances in computer software and hardware,
and with the increasing number of publicly available bioassay
data, virtual screening will continue to remain a vibrant
research field.

Acknowledgements

This work was supported by National Institute of Health (NIH)

grant R01GM069832. The authors are very grateful to all those
who contributed software used in this chapter. We also would like
to thank Alex L. Perryman, Oleg Trott, Ruth Huey, and all mem-
bers of our laboratories for helpful comments and discussions. This
is manuscript # 25035 from The Scripps Research Institute.
250 Sargis Dallakyan and Arthur J. Olson

References

1. Ng R (2008) Drugs: from discovery to approval, 9. Berman HM et al (2002) The protein data bank.
2nd edn. Wiley-Blackwell, Hoboken, NJ Acta Crystallogr D Biol Crystallogr 58:899–907
2. Swinney DC, Anthony J (2011) How were 10. Scior T et al (2012) Recognizing pitfalls in
new medicines discovered? Nat Rev Drug virtual screening: a critical review. J Chem Inf
Discov 10:507–519 Model 52:867–881
3. Jacob RB, Andersen T, McDougal OM 11. Trott O, Olson AJ (2010) AutoDock Vina:
(2012) Accessible high-throughput virtual improving the speed and accuracy of docking with
screening molecular docking software for stu- a new scoring function, efficient optimization, and
dents and educators. PLoS Comput Biol multithreading. J Comput Chem 31:455–461
8:e1002499 12. Rappe AK et al (1992) UFF, a full periodic
4. Lindell SD, Pattenden LC, Shannon J (2009) table force field for molecular mechanics and
Combinatorial chemistry in the agrosciences. molecular dynamics simulations. J Am Chem
Bioorg Med Chem 17:4035–4046 Soc 114:10024–10035
5. Varghese JN et al (1992) The structure of the 13. O'Boyle NM et al (2011) Open Babel: an
complex between influenza virus neuramini- open chemical toolbox. J Cheminform 3:33
dase and sialic acid, the viral receptor. Proteins 14. Perryman AL et al (2010) A dynamic model of
14:327–332 HIV integrase inhibition and drug resistance.
6. McKimm-Breschkin JL (2013) Influenza J Mol Biol 397:600–615
neuraminidase inhibitors: antiviral action and 15. Hanwell MD et al (2012) Avogadro: an
mechanisms of resistance. Influenza Other advanced semantic chemical editor, visualiza-
Respir Viruses 7:25–36 tion, and analysis platform. J Cheminform 4:17
7. Knox C et al (2011) DrugBank 3.0: a compre- 16. Muchmore SW et al (2010) Cheminformatic
hensive resource for ‘omics’ research on drugs. tools for medicinal chemists. J Med Chem
Nucleic Acids Res 39:D1035–D1041 53:4830–4841
8. Li Q et al (2010) PubChem as a public 17. Nicholls A (2008) What do we know and when
resource for drug discovery. Drug Discov do we know it? J Comput Aided Mol Des
Today 15:1052–1057 22:239–255