# Pharmit: Interactive Exploration of Chemical Space

04/19/2016
Abstract

Pharmit (http://pharmit.csb.pitt.edu) provides an online, interactive environment for the virtual screening of large compound databases using pharmacophores, molecular shape, and energy minimization. Users can import, create, and edit virtual screening queries in an interactive browser-based interface. Queries are specified in terms of a pharmacophore, a spatial arrangement of the essential features of an interaction, and molecular shape. Search results can be further ranked and filtered using energy minimization. In addition to a number of pre-built databases of popular compound libraries, users may submit their own compound libraries for screening. Pharmit uses state-of-the-art sub-linear algorithms to provide interactive screening of millions of compounds. Queries typically take a few seconds to a few minutes depending on their complexity. This allows users to iteratively refine their search during a single session. The easy access to large chemical datasets provided by Pharmit simplifies and accelerates structure-based drug design. Pharmit is available under a dual BSD/GPL open-source license.

# Introduction

There are a multitude of software packages and web services that assist in computer aided drug design (Villoutreix 2013), but a relative paucity of web services that support structure-based virtual screening. Those that exist, such as DockBlaster (Irwin 2009), iDrug (Wang 2014), iStar (Li 2014), e-LEA3D (Douguet 2010), and MTiOpenScreen (Labbé 2015), are typically batch-processing services where the user submits a virtual screening job and receives the results hours or days later. They are also usually limited to screening a pre-determined library of compounds of limited size. Alternatively, advanced algorithms enable interactive time-scale searches, but existing web resources (Koes 2012, Koes 2012a) are limited by a single search modality and a restricted search domain. In contrast, Pharmit provides both pharmacophore and molecular shape search modalities as well as ranking of results by energy minimization, and, in addition to providing a variety of pre-built compound libraries, allows users to upload their own compound libraries for screening.

Pharmit takes as its input a predefined pharmacophore query or can elucidate pharmacophore and shape queries from receptor and/or ligand structures. Structures may be provided by the user or extracted directly from the Protein Data Bank (PDB). Pharmacophore and/or molecular shape queries are created and edited in a modern interactive interface powered by 3Dmol.js (Rego 2014), which provides high performance 3D molecular graphics without the need for plugins or Java. Once a query is defined, the user selects and searches a compound library for matching compounds. Results are typically returned in seconds and are displayed in-browser. A variety of filtering and ranking criteria can be applied, and hits can be further refined and ranked using energy minimization. Structure files of the query-optimized hit compounds can be downloaded, and the full session state can be saved and restored. In total, Pharmit provides a comprehensive online platform for structure-based virtual screening.

\label{pharmfig} Pharmacophore as primary query. Each pharmacophore feature has a collapsible menu in the Pharmacophore panel (left) where its type, location, and radius, as well as number of atoms (for hydrophobic features) or directionality (if relevant) can be defined. Selected features are shown as solid spheres and unselected features as meshes. Filters may be set to reduce the number of hits by constraining the number of hits returned for a given conformer or molecule or the overall number of hits. Selecting a hit in the results panel (right) displays it, and its appearance can be adjusted in the visualization filter along with other aspects of the visual display. For example, here the query ligand is shown in light gray, and the selected hit compound is shown in cyan. This query against tyrosine-protein kinase c-Src (PDB 2SRC) is available on the Pharmit Examples page.

# Compound Libraries

Unique to Pharmit is the ability to select from a number of provided compound libraries or to submit a custom library for screening. The library to screen is selected through a pull down menu in the search button (see Figure \ref{pharmfig}).

## Provided Libraries

Large libraries corresponding to compound catalogs from a variety of sources are provided and periodically updated to ensure continued relevance, especially with regard to compound availability from commercial sources. Currently, Pharmit has pre-built libraries generated from CHEMBL21 (Gaulton 2011), with $$>1.4$$ million compounds; ChemDiv (www.chemdiv.com), with $$>1.4$$ million compounds; MolPort (www.molport.com), with $$>6.5$$ million compounds; the NCI Open Chemical Repository (dtp.cancer.gov), with $$>108,000$$ compounds; and PubChem (Kim 2015), with $$>66$$ million compounds. Although a search is limited to the compounds of the selected library, all compounds within these provided libraries are cross-annotated so, for example, it is possible to look up the PubChem record of a compound found by searching the commercial MolPort library to check for known bioactivities.

## Library Creation

Users may submit their own libraries for screening. In the spirit of the open access and open-source nature of Pharmit, users are encouraged to make their submitted libraries publicly accessible, in which case they are available to all users for screening as a user contributed library. However, registered users have the ability to create a private library, as well as remove or update previously submitted libraries.

In order to create a library, compounds may be provided either in the two-dimensional SMILES or three-dimensional SDF formats. If the user uploads compounds in the SMILES format, duplicated canonical SMILES are removed, the molecules are protonated using OpenBabel (OBoyle 2011) using default settings, and only the largest component of a molecule is retained (e.g., salts are removed). Then RDKit (rdkit.org) and the UFF force field (Rappe 1992) are used to generate up to 10 3D conformers for each compound resulting from this procedure. This approach has been shown to generate high quality conformations (Ebejer 2012). Alternatively, if the user provides compounds in the SDF format, the provided structures are assumed to be valid conformers and are used directly, with protonation states determined by OpenBabel.

\label{shapefig} Shape as primary query. The shape menu defines inclusive and exclusive shape extent and tolerance as well as shape query visibility. Hits may be filtered by screening molecular properties relevant to their utility as drugs. Hovering over the name of a hit shows alternate molecular identifiers in all databases. This example uses tyrosine-protein kinase c-Src (PDB ID: 2SRC).

# Query Definition

## Inputs

The typical starting point for a Pharmit session is a ligand-receptor complex structure, although a ligand-only structure or pharmacophore query file may be used as well. A Pharmit session can be automatically initialized using any complex in the PDB by inputing the corresponding PDB accession code on the Pharmit main page and selecting how active site water molecules should be treated (ignored, as part of the receptor, or as part of the ligand). The dropdown menu next to the box where a PDB code may be entered allows the user to select which ligand found in the PDB file should be used as the basis of the query.

Alternatively, a user can upload their own complex, in which case the receptor and ligand structures must be in separate files; these are uploaded by clicking ‘Enter Pharmit Search’ on the main page and then choosing ‘Load receptor’ and ‘Load features’ on the resulting page. Any file format supported by OpenBabel may be used. Note that the query ligand must be pre-positioned within the binding site of the receptor - Pharmit does not perform docking. Pharmit prepares the receptor by protonating it with OpenBabel, but no other modifications are made. Thus the user must decide whether there are missing residues that should be included, if the histidine protonation state is correct, or if any other structural changes to the receptor are necessary.

Pharmit can also accept initiate queries without a receptor, either in the form of a ligand structure or an externally generated pharmacophore query file, but in this case features that require a receptor, such as energy minimization, will not be available.

## Pharmacophore Queries

A pharmacophore (Koes 2015a, Yang 2010, Leach 2010) defines the essential features of an interaction. Importantly, a pharmacophore includes the spatial arrangement of these features. Features supported by Pharmit include hydrogen bond acceptors and donors, negative and positive charges, aromatics, and hydrophobic features. As shown in Figure \ref{pharmfig}, a pharmacophore query specifies these features using tolerance spheres. Compounds match if they can be positioned so that their corresponding features are located within these spheres. Some features can have additional constraints, such as size (number of atoms) for hydrophobic features and direction for hydrogen bonds and aromatics.

Pharmit will identify all pharmacophore features present in a provided ligand structure. If a receptor structure is provided, it will identify which of these features are relevant to the protein-ligand interaction using distance cutoffs between corresponding features on the receptor and ligand (e.g., a hydrogen donor on the ligand and acceptor on the receptor). Only the interacting features will be enabled. Alternatively, a pharmacophore query can be initialized using pharmacophore files in MOE (Chemical Computing Group 2016), LigBuilder (Wang 2000), LigandScout (Wolber 2005), PharmaGist (Schneidman-Duhovny 2008) or Pharmer (Koes 2011) query formats. The features of the query can be interactively edited within the Pharmit interface as changes to the query editor are immediately reflected in the molecular viewer and clicking on a feature in the viewer selects it for editing, as shown in Figure \ref{pharmfig}.

## Shape Queries

Similarity of molecular shape is a common method of structure-based virtual screening (Nicholls 2010). Pharmit uses the Volumetric Aligned Molecular Shapes (VAMS) (Koes 2014) method of shape search, which uses inclusive and exclusive shape constraints to identify matching molecular shapes. In Pharmit, inclusive constraints are specified using the shape of the provided ligand or by manually specified inclusion spheres. Inclusive constraints specify a minimum bound on the desired molecular shape; matching compounds will overlap these constraints. Exclusive constraints are specified using the shape of the provided receptor or by manually specified exclusion spheres. Exclusive constraints are used to limit the desired molecular shape; matching compounds are prohibited from overlapping these constraints. Both constraints are represented using voxelized volumes, as shown in Figure \ref{shapefig}, and can be adjusted by growing or shrinking the constraint volume. A shape-first search with pharmacophore restraints is distinct from the commonly used ROCS with color shape method (Hawkins 2007) as it does not optimize the position with respect to the pharmacophores.

\label{minfig} (Left) An example of a pharmacophore-aligned hit from the DUDe SRC target dataset for a query derived from PDB 2SRC, and (Right) the same example after energy minimization. Steric clashes in the back pocket are resolved by rotating a phenyl resulting in an energy score of -12.46 kcal/mol.

# Visualization

The visualization scheme of the Pharmit interface is highly customizable. Both the color and representation style (e.g., sticks, spheres, cartoon, etc.) of the query receptor and ligand and the result compounds can be set individually as demonstrated in Figure \ref{pharmfig}. The transparency of the receptor surface, which is shown colored by electrostatic potential, may be adjusted from not present to fully opaque, and the background may be toggled between white and black. To facilitate inspection of molecular models, the query and results sidebars may be hidden to provide a full-window molecular viewer. Manipulation of the 3D viewer will be familiar to users of PyMol (Schrödinger, LLC 2015): left mouse button to rotate, right or scroll wheel to zoom, and middle button to translate.

# Search

Pharmit provides two search modalities depending on whether the pharmacophore or shape query is the primary query. In both cases, the primary query determines the pose alignment of the hit compounds and the secondary query serves as an additional filter. If desired, the complete set of query-aligned results may be saved as a compressed SDF structure file.

## Pharmacophore/Shape Search

If the primary query is the pharmacophore query, the selected database is searched for compounds that match the specified pharmacophore using the Pharmer (Koes 2011) search technology. Pharmer has comparable virtual screening performance to other pharmacophore methods (Sanders 2012) but is orders of magnitude faster.

Results are aligned to the pharmacophore to minimize the root mean squared deviation (RMSD) between the query features and the hit compound features. Results are sorted with respect to this pharmacophore RMSD. An example of a pharmacophore-oriented search is shown in Figure \ref{pharmfig}.

If a shape query is also present, then the shape constraints are applied to the pharmacophore-aligned pose; only heavy atom centers are compared to the shape constraints. That is, the pharmacophore-aligned results are filtered to ensure that hits have at least one heavy atom center that falls within the inclusive shape and no heavy atom center that falls within the exclusive shape. With this modality, the exclusive shape is generally the most useful, as it provides a way to eliminate compounds that match the pharmacophore but have significant steric clashes with the receptor. Due to the importance and increased specificity of the interactions specified by a pharmacophore, pharmacophore search followed by shape filtering is the default and recommended search modality.

## Shape/Pharmacophore Search

If the primary query is the shape query, VAMS (Koes 2014) is used to search a shape index (Koes 2015) for matches. Although VAMS does not exhibit as good virtual screening performance as more expensive methods, such as ROCS (Hawkins 2007), it is orders of magnitude faster and serves as an effective shape-based pre-screen (Koes 2014). With VAMS, molecules are pre-aligned to their moments of inertia and hits are aligned to the moments of inertia of the query. Molecular shapes are computed using the solvent excluded volume and are stored at a 0.5Å resolution. A shape matches the query if, in its aligned position, the entirety of the inclusive shape constraint is contained within the shape while no part of the shape overlaps the exclusive shape. This is a more stringent requirement than that which is imposed when the shape query is used as a filter to a pharmacophore search. Matching compounds are sorted with respect to their Tanimoto shape similarity with the query ligand. An example of a shape-oriented search is shown in Figure \ref{shapefig}.

If a pharmacophore query is present, it is used to filter the shape-aligned results. In this case, the pharmacophore features of the hit compound must fall within the tolerance spheres of the pharamacophore query with the compound in the shape-aligned pose. That is, a compound that would match a pharmacophore query if the query were the primary query may still be filtered out.

Pharmit allows users to specify additional, non-structural filters on the results. As shown in Figure \ref{pharmfig}, the number of compounds returned can be reduced by restricting the number poses returned for every conformation, the number of conformations returned for every compound, and the total number of compounds.

Compounds may be filtered by specifying desired ranges for molecular properties that are commonly used to identify drug-like molecules. As shown in Figure \ref{shapefig}, these include molecular weight, the number of rotatable bonds, logP (a measure of lipophilicity), topological polar surface area (indicative of the compound’s ability to permeate cell membranes), the number of aromatic groups, the number of hydrogen bond acceptors, and the number of hydrogen bond donors. Properties are precomputed using OpenBabel (OBoyle 2011).

## Result Browser

When a search is initiated, a results sidebar opens to display the search results, as seen in Figures \ref{pharmfig} and \ref{shapefig}. Results may be sorted in increasing or decreasing order based on RMSD (for pharmacophore searches), similarity score (for shape searches), molecular weight, or number of rotatable bonds. Clicking on a result displays the query-aligned compound in the molecular viewer, as Figure \ref{pharmfig} shows. If desired, a new Pharmit session can be initiated from the ligand-protein complex represented by the selected hit compound. Compounds are annotated with all known database identifiers, as shown in Figure \ref{shapefig}. Where applicable, these identifiers are linked to the source database, so it is possible to immediately investigate bioactivity data in PubChem or bring up the compound’s order page from a commercial vendor.

# Energy Minimization

The query-aligned poses returned by pharmacophore and shape search are fully determined by the query. They are not influenced by interactions with the receptor. Furthermore, although the pose of the result is optimized to match the query, the conformation of the result is taken directly from the pre-generated conformers of the search database. Energy minimization of results optimizes both the pose and conformation of identified hits with respect to the provided receptor using the AutoDock Vina (Trott 2010) scoring function and smina (Koes 2013), a fork of AutoDock Vina with enhanced minimization functionality.

Minimized compounds adopt a conformation at a local minimum identified by performing gradient descent on the energy surface starting from the initial query aligned pose. Results are sorted according to predicted binding affinity in kcal/mol (more negative values are more favorable). Also reported is a minimized RMSD (mRMSD) that is the RMSD between the query-aligned pose and the minimized pose. This provides an indication of how far the compound has deviated from the original query. Minimized results may be sorted and filtered by affinity and mRMSD to eliminate poses with unfavorable binding energies (e.g., $$>-6$$ kcal/mol) and significant deviations from the original query (e.g. 2Å). An example of an energy minimized compound is shown in Figure \ref{minfig} and contrasted with the unminimized pharmacophore aligned pose. The complete set of minimized compound structures can be saved, including scoring annotations, as a compressed SDF structure file.

# Example

To demonstrate Pharmit’s capabilities and typical usage, this section describes a virtual screen of Tyrosin-protein kinase C-SRC based off a complex (PDB 2SRC) with a nonhydrolyzable ATP analog (ANP). The resulting query is available as an interactive example from the Pharmit examples page. For this example we submitted the compounds of the the DUDe (Mysinger 2012) SRC target benchmark as a contributed library. Compounds were renamed to include the keyword ‘active’ if they were active compounds, which allows Pharmit to automatically compute the enrichment factor (EF) and F1 score (geometric mean of recall and precision) of a search.

Beginning on the Pharmit main page, the user initiates a search by typing ‘2SRC’ into the ‘start from PDB’ box. This will retrieve the ligand names, ANP and PTR, from the PDB file, and they are displayed in the dropdown menu next to the PDB code box. The user then selects the ANP ligand and, for this example, chooses to ignore the binding site waters. After clicking ‘submit,’ the user will be taken to the main Pharmit interface. A set of interacting pharmacophore features will be automatically generated from the protein-ligand complex. The user then selects the DUDe SRC Benchmark from the Contributed Libraries list in the search selection menu. This contains 514,797 conformers of 35,024 molecules, of which 524 are known actives.

Pharmit identifies 26 pharmacophore features in ANP, 14 of which are interacting with the receptor. Searching with this default query yields no hits as it is overly specific. In general, we find that queries rarely benefit from having more than 5 or 6 distinct features. To generate a useful query for 2SRC, we rationally target the canonical hinge site, with which ANP strongly interacts, by including a hydrogen donor and acceptor and an aromatic ring to mimic the adenine moiety. Searching with this reduced query produces 92,398 hits (matching conformations) with an EF of 2.8. Adding a directionality constraint to the aromatic ring reduces the number of hits by two thousand and slightly increases the EF. The user can then interactively explore adding additional interacting features to the query as each query takes at most a few seconds. The addition of a hydrogen acceptor interaction on the ribose of the ligand generates 1,881 hits with an EF of 16.8 and produces the pharmacophore shown in Figure \ref{pharmfig}. The addition of an exclusive shape constraint with a tolerance of 2.5, which filters out compounds with severe steric clashes with the receptor, decreases the number of hits to 1,248 and increases the EF to 18.4 as shown in Figure \ref{minfig}. Alternative queries can yield even higher enrichment factors at the expense of lower F1 scores due to fewer total active compounds being returned. In this case the user is guided by a benchmark library, but a similar interactive exploration is possible through rational investigation of the hit compounds generated by a query. Query results can be minimized to produce a more meaningful ranking, as shown in Figure \ref{minfig}, where the top 13 compounds, and 16 of the top 20, are correctly identified actives.

# Discussion

Pharmit represents a significant advance over existing browser-based virtual screening servers. Among its virtues are its search speed, which is on the order of seconds; its high-resolution, smoothly animated visual interface; its customizability, including the capability to create custom compound libraries; and its expansive suite of search features, including an array of chemical constraints that allow the user to rapidly identify hits that are also promising drug candidates. It democratizes structure-based computer-aided drug discovery by offering free and open access to state-of-the-art software for exploring chemical space, making it ideal for research and education alike. Educators can deploy a high quality product to illustrate fundamental concepts in physical chemistry. Researchers can adapt the project to incorporate it into their existing drug discovery workflows. In settings where intellectual property protection is an issue, the entirely open source nature of Pharmit allows it to be deployed locally by the user, ensuring the complete privacy of queries and results. Pharmit anticipates the needs of the medicinal chemist by facilitating direct comparison between a search ligand and hits, permitting in-browser energy minimization, and enabling the user to rapidly launch new sessions using molecules found via a query. Pharmit enables the user to perform structure-based virtual screening in a manner that is fast and intuitive.

# Funding

This work is funded through R01GM108340 from the National Institute of General Medical Sciences. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of General Medical Sciences or the National Institutes of Health.

# Acknowledgements

We thank Charles Yuan for assistance with web site design and the many early users of Pharmit for their helpful feedback.

### References

1. Bruno O. Villoutreix, David Lagorce, Céline M. Labbé, Olivier Sperandio, Maria A. Miteva. One hundred thousand mouse clicks down the road: selected online resources supporting drug discovery collected over a decade. Drug Discovery Today 18, 1081–1089 Elsevier BV, 2013. Link

2. John J. Irwin, Brian K. Shoichet, Michael M. Mysinger, Niu Huang, Francesco Colizzi, Pascal Wassam, Yiqun Cao. Automated Docking Screens: A Feasibility Study. J. Med. Chem. 52, 5712–5720 American Chemical Society (ACS), 2009. Link

3. Xia Wang, Haipeng Chen, Feng Yang, Jiayu Gong, Shiliang Li, Jianfeng Pei, Xiaofeng Liu, Hualiang Jiang, Luhua Lai, Honglin Li. iDrug: a web-accessible and interactive drug discovery and design platform. Journal of Cheminformatics 6, 28 Springer Science and Business Media, 2014. Link

4. Hongjian Li, Kwong-Sak Leung, Pedro J. Ballester, Man-Hon Wong. istar: A Web Platform for Large-Scale Protein-Ligand Docking. PLoS ONE 9, e85678 Public Library of Science (PLoS), 2014. Link

5. D. Douguet. e-LEA3D: a computational-aided drug design web server. Nucleic Acids Research 38, W615–W621 Oxford University Press (OUP), 2010. Link

6. Céline M. Labbé, Julien Rey, David Lagorce, Marek Vavruša, Jérome Becot, Olivier Sperandio, Bruno O. Villoutreix, Pierre Tufféry, Maria A. Miteva. MTiOpenScreen: a web server for structure-based virtual screening. Nucleic Acids Res 43, W448–W454 Oxford University Press (OUP), 2015. Link

7. David Koes, Kareem Khoury, Yijun Huang, Wei Wang, Michal Bista, Grzegorz M. Popowicz, Siglinde Wolf, Tad A. Holak, Alexander Dömling, Carlos J. Camacho. Enabling Large-Scale Design Synthesis and Validation of Small Molecule Protein-Protein Antagonists. PLoS ONE 7, e32839 Public Library of Science (PLoS), 2012. Link

8. D. R. Koes, C. J. Camacho. ZINCPharmer: pharmacophore search of the ZINC database. Nucleic Acids Research 40, W409–W414 Oxford University Press (OUP), 2012. Link

9. N. Rego, D. Koes. 3Dmol.js: molecular visualization with WebGL. Bioinformatics 31, 1322–1324 Oxford University Press (OUP), 2014. Link

10. A. Gaulton, L. J. Bellis, A. P. Bento, J. Chambers, M. Davies, A. Hersey, Y. Light, S. McGlinchey, D. Michalovich, B. Al-Lazikani, J. P. Overington. ChEMBL: a large-scale bioactivity database for drug discovery. Nucleic Acids Research 40, D1100–D1107 Oxford University Press (OUP), 2011. Link

11. Sunghwan Kim, Paul A. Thiessen, Evan E. Bolton, Jie Chen, Gang Fu, Asta Gindulyte, Lianyi Han, Jane He, Siqian He, Benjamin A. Shoemaker, Jiyao Wang, Bo Yu, Jian Zhang, Stephen H. Bryant. PubChem Substance and Compound databases. Nucleic Acids Res 44, D1202–D1213 Oxford University Press (OUP), 2015. Link

12. Noel M OBoyle, Michael Banck, Craig A James, Chris Morley, Tim Vandermeersch, Geoffrey R Hutchison. Open Babel: An open chemical toolbox. Journal of Cheminformatics 3, 33 Springer Science $$\mathplus$$ Business Media, 2011. Link

13. A. K. Rappe, C. J. Casewit, K. S. Colwell, W. A. Goddard, W. M. Skiff. UFF a full periodic table force field for molecular mechanics and molecular dynamics simulations. J. Am. Chem. Soc. 114, 10024–10035 American Chemical Society (ACS), 1992. Link

14. Jean-Paul Ebejer, Garrett M. Morris, Charlotte M. Deane. Freely Available Conformer Generation Methods: How Good Are They?. Journal of Chemical Information and Modeling 52, 1146–1158 American Chemical Society (ACS), 2012. Link

15. David Ryan Koes. Pharmacophore Modeling: Methods and Applications. In Methods in Pharmacology and Toxicology. Springer Science and Business Media, 2015. Link

16. Sheng-Yong Yang. Pharmacophore modeling and applications in drug discovery: challenges and recent advances. Drug Discovery Today 15, 444–450 Elsevier BV, 2010. Link

17. Andrew R. Leach, Valerie J. Gillet, Richard A. Lewis, Robin Taylor. Three-Dimensional Pharmacophore Methods in Drug Discovery. J. Med. Chem. 53, 539–558 American Chemical Society (ACS), 2010. Link

18. Chemical Computing Group. Molecular Operating Environment (MOE). Chemical Computing Group Inc. Montreal, QC, 2016.

19. Renxiao Wang, Ying Gao, Luhua Lai. LigBuilder: A Multi-Purpose Program for Structure-Based Drug Design. Journal of Molecular Modeling 6, 498–516 Springer Science $$\mathplus$$ Business Media, 2000. Link

20. Gerhard Wolber, Thierry Langer. LigandScout:  3-D Pharmacophores Derived from Protein-Bound Ligands and Their Use as Virtual Screening Filters. Journal of Chemical Information and Modeling 45, 160–169 American Chemical Society (ACS), 2005. Link

21. D. Schneidman-Duhovny, O. Dror, Y. Inbar, R. Nussinov, H. J. Wolfson. PharmaGist: a webserver for ligand-based pharmacophore detection. Nucleic Acids Research 36, W223–W228 Oxford University Press (OUP), 2008. Link

22. David Ryan Koes, Carlos J. Camacho. Pharmer: Efficient and Exact Pharmacophore Search. Journal of Chemical Information and Modeling 51, 1307–1314 American Chemical Society (ACS), 2011. Link

23. Anthony Nicholls, Georgia B. McGaughey, Robert P. Sheridan, Andrew C. Good, Gregory Warren, Magali Mathieu, Steven W. Muchmore, Scott P. Brown, J. Andrew Grant, James A. Haigh, Neysa Nevins, Ajay N. Jain, Brian Kelley. Molecular Shape and Medicinal Chemistry: A Perspective. J. Med. Chem. 53, 3862–3886 American Chemical Society (ACS), 2010. Link

24. DR Koes, CJ Camacho. Shape-based virtual screening with volumetric aligned molecular shapes.. J Comput Chem 35, 1824-34 (2014).

25. Paul C. D. Hawkins, A. Geoffrey Skillman, Anthony Nicholls. Comparison of Shape-Matching and Docking as Virtual Screening Tools. J. Med. Chem. 50, 74–82 American Chemical Society (ACS), 2007. Link

26. Schrödinger, LLC. The PyMOL Molecular Graphics System, Version 1.8. (2015).

27. Marijn P. A. Sanders, Arménio J. M. Barbosa, Barbara Zarzycka, Gerry A.F. Nicolaes, Jan P.G. Klomp, Jacob de Vlieg, Alberto Del Rio. Comparative Analysis of Pharmacophore Screening Tools. Journal of Chemical Information and Modeling 52, 1607–1620 American Chemical Society (ACS), 2012. Link

28. DR Koes, CJ Camacho. Indexing Volumetric Shapes with Matching and Packing.. Knowl Inf Syst 43, 157-180 (2015).

29. Oleg Trott, Arthur J. Olson. AutoDock Vina: Improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading. Journal of Computational Chemistry 31, 455–461 Wiley Subscription Services, Inc., A Wiley Company, 2010. Link

30. DR Koes, MP Baumgartner, CJ Camacho. Lessons learned in empirical scoring with smina from the CSAR 2011 benchmarking exercise.. J Chem Inf Model 53, 1893-904 (2013).

31. Michael M. Mysinger, Michael Carchia, John. J. Irwin, Brian K. Shoichet. Directory of Useful Decoys Enhanced (DUD-E): Better Ligands and Decoys for Better Benchmarking. J. Med. Chem. 55, 6582–6594 American Chemical Society (ACS), 2012. Link