A simple method for finding a protein’s ligand-binding pockets
© Saberi Fathi and Tuszynski; licensee BioMed Central Ltd. 2014
Received: 9 January 2014
Accepted: 11 July 2014
Published: 19 July 2014
This paper provides a simple and rapid method for a protein-clustering strategy. The basic idea implemented here is to use computational geometry methods to predict and characterize ligand-binding pockets of a given protein structure. In addition to geometrical characteristics of the protein structure, we consider some simple biochemical properties that help recognize the best candidates for pockets in a protein’s active site.
Our results are shown to produce good agreement with known empirical results.
The method presented in this paper is a low-cost rapid computational method that could be used to classify proteins and other biomolecules, and furthermore could be useful in reducing the cost and time of drug discovery.
Essential information regarding protein function is generally dependent on the protein’s tertiary structure. This includes the enzymatic function of a protein, and also the binding of ligands, such as small molecule inhibitors . Methods developed for predicting an enzymatic function of a protein by identifying catalytic residues include: finding local characteristics of functional residues [2, 3], applying known templates of active sites [4, 5] or identifying the surface shape of active sites [6–10].
In order to predict ligand binding (sites, poses and affinities), we first need to determine a 3-dimensional structure of the protein in question, which can be done using several experimental or computational methods [11, 12]. Structure-based pocket prediction employs geometrical algorithms or probes mapping/docking algorithms . Comparing these two kinds of methods, it can be said that the geometrical algorithms have low computational costs in contrast to the mapping/docking and scoring of molecular fragments, but the latter algorithms have a greater physical meaning. Geometrical algorithms analyze protein surfaces, and once a structure has been determined, a number of algorithms may be used to predict binding pockets on the protein surface [14–19]. One such example, SURFNET , fits spheres into the spaces between protein atoms and finds gap regions. The results obtained this way correspond to the cavities and keys of a given protein. An algorithm based on geometric hashing called VISGRID  uses the visibility of constituent atoms to identify cavities. “Active site points” are identified by PASS . In this method the protein surface is coated with a layer of spherical probes, then those that clash with the protein or which are not sufficiently buried are filtered out. The active site points are identified from the final probes. Another method is LIGSITE [14, 21], which is an improvement of the POCKET algorithm . This algorithm puts protein-occupied space in a grid and identifies clefts by scanning areas that are enclosed on both sides by the protein’s atoms. An alpha-shape algorithm is used by CAST  and APROPOS . DRUGSITE  and POCKET-FINDER , in addition to the protein’s shape, also consider physicochemical properties for identification of ligand binding pockets. Further geometrical algorithms are TRAVEL DEPTH , VOIDOO , and CAVITY SEARCH . QSITEFINDER  uses interaction energy computation between the protein and a van der Waals probe to find favorable binding sites. Some methods using mapping/docking and scoring of molecular fragment concepts are described by Dennis et al. , Kortvelyesi et al. , Ruppert et al. , and Verdonk et al. . There are also several docking based methods that use ligands to probe the proteins for binding sites [31–34].
Computer-aided drug design often applies protein–ligand docking methods, most commonly structure-based methods. These methods provide support to the rational design and optimization of novel drug candidates . Many structure-based protein–ligand docking methods have been reported in the literature [36–41]. These methods generally rely on first identifying a ligand-binding pocket in the protein structure.
Finding a comprehensive, fast and automated method that can accurately predict ligand-binding pockets on protein surfaces is a major challenge in virtual screening biophysics. This goal leads us to introduce a new method for finding putative ligand-binding pockets on a protein surface, and for identifying the most important characteristics of these pockets: surface area, volume, and potential interacting atoms. This information could be used to cluster protein pockets into similarity classes, and could be a valuable resource leading to a significant decrease in the cost and time expended in the drug discovery process.
The method we present in this paper is based on computational geometry and voxelization concepts. In this method we do not use Delaunay tessellation, the vision criterion, or fitting spheres between atoms, in contrast to some of the methods mentioned above. The CASTp method has used the Delaunay triangulation and the Voronoi concepts to find putative pockets and voids. This method triangulates the surface atoms and clusters triangles by merging small triangles to neighboring large triangles [14, 17]. In our work we simply use the convex hull concept and generate a pocket by a grid box formed by the extreme points of a triangle. Then, we consider only the atoms closest to the triangle in the formed pocket. The distance to the convex hull is used for choosing the surface atoms. Thus, our method is not iterative and does not require a flow through all points, hence the computational cost is relatively low. We also take only a given number of empty voxel neighbors for each atom. Voxelization of space for finding putative pockets does not have an essential role for finding surface atoms, unlike VISGRID or grid-based methods, which are based on searching for empty voxels in different directions. We also use voxelization for finding the positions of possible ligands and also to determine physical properties of the pockets.
Comparative modeling methods use fold assignment and template selection for comparing the target protein to a set of proteins with known structures and to search for homologous proteins that have approximately similar structures. Some of these methods are BLAST [42, 43], PSI-BLAST  and HHpred . I-TASSER  is a composite approach of comparative modeling and threading methods . A summary of comparative modeling is given by . In our method we also consider some biochemical properties of the protein’s atoms and residues as is explained below. Hence, the proposed method is not purely geometrical. We demonstrate that the results obtained using this method are in good agreement with empirically known results. Hence developing it further may offer even more accurate and reliable results.
If all atoms contained in a set of the pocket atoms exist in the other pocket, it has an overlap of 100%. However, the second pocket may have more atoms than the first one, i.e. it has all atoms of the first pocket plus other atoms. For example, the overlap between pockets #1 and #2 might be 100% while the overlap between pockets #2 and #1 is only 50%, because the number of atoms in pocket #2 is twice as large as the number of atoms in pocket #1, and all atoms belonging to pocket #1 are also contained in pocket #2, but only a half of the atoms in pocket #2 are also in pocket #1. Accumulating all pockets with a given overlap between them as new pockets is the next step.
HBD: OG1 (OH)
HBD: OG (OH)
HBA: NE2 (NH2)
HBA: ND2 (NH2)
HBA: O – HBD: N, OH – CR: CE1, CE2, CD1, CD2, CZ, CG
Sul: SG (SH)
vdW: CE (CH3) – Sul: SD (S-CH3)
vdW: CB (CH3)
vdW: CB (CH2), CD (CH2), CG (CH2)
vdW: CD1 (CH3), CD2 (CH3), CG (CH)
vdW: CG1 (CH3), CG2 (CH3), CB (CH)
vdW: CD1 (CH3)
HBA: OD1(C = O) – Ion(−): OD2 (OH)
HBA: OE1(C = O) – Ion(−): OE2 (OH)
Ion(+): NZ (NH3)
Ion(+): NH1 (NH2) trans, NH2 (NH2) cis
Ion(+): NE1 (NH2) trans, NE2 (NH2) cis – CR: CD1, CE1, CD2, CE2, CG
CR: CG, CD1, CE1, CZ, CE2, CD2
HBD: NE1 (NH) – CR: CD2, CE2, CZ2, CH2, CZ3, CE3
HBD: OH – CR: CD1, CE1, CE2, CZ, CD2, CG
C-Ring in ligand
C or N atoms in ligand recognizing by connection information in the PDB
Unprotonated atoms in ligand
1) O has a connection with N, P or Zn
2) O only has a connection with C
Protonated atoms in ligand
2) N has only two connection with C
A detailed description of the algorithm is given in the following:
Input protein atom position data, and define a box by using the extreme positions of the atoms.
Voxelize the box by considering the voxel with 1 Å in length, width and height.
Compute the convex hull surrounding the protein atoms and obtain the volume of the convex hull and the surface area of atoms.
Separate empty voxels (possible ligand atom positions) from voxels filled by the protein atoms in the convex hull.
Define the pockets by the volume generated by the vertices of each triangle on the convex hull.
Compute the overlap between two neighboring pockets and assemble the pockets with an overlap greater than a minimum value (reconstruct new pockets).
Find the physical properties of the pockets such as depth, surface and volume.
Find the residues corresponding to the pocket atoms.
Assess the biochemical conditions [49, 50] as introduced in Table 1 (we use the IUPAC nomenclature  and the PDB format ). In this step we can find the atoms and residues that participate in the potential active site.
Compare physical and biochemical properties between ligand atoms (Table 2) and the atoms of a given pocket, such as: the size of pockets (depth, surface and volume) with ligand size, the number of hydrogen donor/acceptor atoms, possible rings, or van der Waals interactions, etc.
- 11.Compute the number of correct residues predicted in each pocket of the unliganded protein and divide it by the number of residues in an “active site” of the liganded protein as reported in the PDB, i.e.
Optional step. Compute the minimum distance between the ligand atoms and each residue atoms in the pocket. Then, filter residues of a pocket with the minimum distance greater than the given values, for example 3.50 Å.In Figure 3, we illustrate these steps in 2-dimensional space for better clarity. Here, we need to use a line instead of a triangle to define a pocket. Figure 5 uses the example of the protein labeled 1A6U in the PDB. It shows 3-dimensional atomic positions of the protein and the atoms that belong to a pocket.
Results and discussion
In reality, the geometrical criteria give initial information about physical properties for the possible protein-protein or protein-ligand docking, determining shapes, sizes, etc. For docking to occur, the recognized geometrical protein pocket should be a protein’s active site. Finding active sites is very complicated for both in vitro and in silico methods. There are many computer programs that find active sites [13–23] but they have high computational cost associated with them and also they do not typically determine physical properties of the active site which means that we need to find a ligand in spite of lacking some important information. Therefore, it is imperative to use mixed geometrical and biochemical methods to find possible pockets in a protein. This paper has introduced a method to find protein pockets with a higher probability of interactions than based on exclusively biochemical methods. This method offers a speed-up of the drug discovery process by allowing clustering of both the protein pockets and ligands.
Pockets and their characteristics recognized by our method for 1A6U protein atoms
Num. of Atoms
Num. of Empty voxels
Surface of Pocket
Depth of Pocket
NoA**HA a Bonds
NoA HD b Bonds
NoA Ionic Bonds
NoA Sulfur Bonds
of the 1stAS, HAP c
of the 2ndAS, AC1 c
1A6U best pockets with residues in common with the 2 active sites, HAP and AC1
POCKET # 1, cf = 0.31 & 0.33
ASN 354H (11.61)
SER 331H (10.79)
TYR 34 L (4.27)
ASP 352H (7.07)
THR 328H (14.41)
TYR 332H (8.34)
ILE 351H (6.25)
THR 330H (12.29)
TYR 401H (2.92)
SER 32 L (6.81)
TRP 333H (1.734)
TYR 402H (5.75)
POCKET # 39, cf = 0.31 & 0.22
ALA 2 L (15.1365)
HIS 97 L (6.8477)
THR 26 L (15.7431)
ARG 350H (2.89)
ILE 348H (9.34)
TRP 98 L (3.24)
ASN 96 L (7.12)
LYS 359H (5.38)
TRP 347H (4.78)
ASN 361H (9.75)
LYS 365H (14.84)
TYR 94 L (7.84)
GLU 362H (12.30)
PHE 364H (13.46)
TYR 360H (8.34)
GLY 349H (6.45)
SER 366H (17.38)
VAL 99 L (9.69)
POCKET # 137, cf = 0.25 & 0.33
ASP 400H (5.44)
THR 31 L (8.29)
TYR 401H (2.92)
SER 405H (3.65)
TYR 34 L (4.27)
TYR 402H (5.75)
POCKET # 143, cf = 0.44 & 0.56
ARG 350H (2.89)
SER 95 L (5.42)
TYR 332H (8.34)
ASN 354H (11.61)
SER 331H (10.79)
TYR 401H (2.92)
ASP 352H (7.07)
TRP 93 L (3.36)
TYR 402H (5.75)
ILE 351H (6.25)
TRP 333H (1.73)
SER 32 L (6.81)
TYR 34 L (4.27)
Table 3 gives all pockets of 1A6U, where only the two last columns are obtained by the comparison of the results with the binding sites HAP and AC1 of 1A6W (the corresponding liganded protein of 1A6U). In Table 3 the pockets are numbered and ordered arbitrarily. This table and all results were produced independently of the final answer.
For illustration purposes we have taken the set of 48 and 86 “liganded and unliganded proteins”, respectively, listed in the supplementary material of Li et al.  and downloaded the files from the PDB site (see Additional file 1 for a list of the PDB files). We found the pockets of the unliganded proteins, and then we compared these pockets with the known active sites reported in the PDB files of the corresponding liganded proteins.
Performance comparison of our results with the other methods CASTp, LIGSITE, PASS, SURFNET and VISGRID
48 Unbound structures
86 Unbound structures
VISGRID: Top 0.8% voxels
Our method: Overlap 0.8
An important step which allows a decrease of the time and effort for the drug discovery process is to find suitable ligands through in silico methods using, for example, the virtual screening techniques. Our algorithm is easy to use and the cost of computation is approximately between 10 seconds for small proteins and up to 320 seconds for large proteins. The program was implemented in Matlab. The computer used for these computations is a laptop with an Intel Core i7 CPU and 8 GB RAM. The program usually uses 13% of the CPU time, but sometimes for a while it uses up to 50%. The program also while occupied in computation usually required less than 0.5 GB of RAM memory, but it was observed for some proteins to go up to 2 GB. The execution time for the 130 pair dataset is given in Additional file 3.
In this paper, we have introduced a new simple method for predicting putative ligand-binding protein pockets. For each pocket, we can identify possible interacting protein atoms and residues, surface atoms, and also determine the size of a pocket (volume, surface area and depth). This information can help us verify possible ligands having a shape and size that is geometrically compatible with the pocket, and thus could be docked to the protein. We have used some biochemical properties to find the possible interacting atoms and residues in the pockets. Our method is a low cost computational method which voxelizes the protein space, and uses the convex hull concept commonly employed in computational geometry. This method could be used to classify proteins by the geometric properties of their pockets and also by their biochemical properties. An application of this method could be useful in reducing the cost and time of drug discovery.
SMSF acknowledges grant number 2/21897 from Ferdowsi University of Mashhad. JAT gratefully acknowledges research support received from the National Science and Engineering Research Council of Canada, the Canadian Breast Cancer Foundation, and the Allard Foundation.
- Polgár L: The catalytic triad of serine peptidases. Cell Mol Life Sci CMLS 2005, 62: 2161–2172.View ArticlePubMedGoogle Scholar
- Mooney SD, Liang MH-P, DeConde R, Altman RB: Structural characterization of proteins using residue environments. Proteins 2005, 61: 741–747.PubMed CentralView ArticlePubMedGoogle Scholar
- Shulman-Peleg A, Nussinov R, Wolfson HJ: Recognition of functional sites in protein structures. J Mol Biol 2004, 339: 607–633.View ArticlePubMedGoogle Scholar
- Fetrow JS, Godzik A, Skolnick J: Functional analysis of the Escherichia coli genome using the sequence-to-structure-to-function paradigm: identification of proteins exhibiting the glutaredoxin/thioredoxin disulfide oxidoreductase activity. J Mol Biol 1998, 282: 703–711.View ArticlePubMedGoogle Scholar
- Wallace AC, Borkakoti N, Thornton JM: TESS: a geometric hashing algorithm for deriving 3D coordinate templates for searching structural databases. Application to enzyme active sites. Protein Sci Publ Protein Soc 1997, 6: 2308–2323.View ArticleGoogle Scholar
- Connolly ML: Solvent-accessible surfaces of proteins and nucleic acids. Science 1983, 221: 709–713.View ArticlePubMedGoogle Scholar
- Goldman BB, Wipke WT: QSD quadratic shape descriptors. 2. Molecular docking using quadratic shape descriptors (QSDock). Proteins 2000, 38: 79–94.View ArticlePubMedGoogle Scholar
- Duncan BS, Olson AJ: Approximation and characterization of molecular surfaces. Biopolymers 1993, 33: 219–229.View ArticlePubMedGoogle Scholar
- Exner TE, Keil M, Brickmann J: Pattern recognition strategies for molecular surfaces. I. Pattern generation using fuzzy set theory. J Comput Chem 2002, 23: 1176–1187.View ArticlePubMedGoogle Scholar
- Kinoshita K, Nakamura H: Identification of protein biochemical functions by similarity search using the molecular surface database eF-site. Protein Sci Publ Protein Soc 2003, 12: 1589–1595.View ArticleGoogle Scholar
- Rupp B, Wang J: Predictive models for protein crystallization. Methods San Diego Calif 2004, 34: 390–407.View ArticleGoogle Scholar
- Arnold K, Bordoli L, Kopp J, Schwede T: The SWISS-MODEL workspace: a web-based environment for protein structure homology modelling. Bioinforma Oxf Engl 2006, 22: 195–201.View ArticleGoogle Scholar
- An J, Totrov M, Abagyan R: Comprehensive identification of “druggable” protein ligand binding sites. Genome Inform Int Conf Genome Inform 2004, 15: 31–41.Google Scholar
- Huang B, Schroeder M: LIGSITEcsc: predicting ligand binding sites using the connolly surface and degree of conservation. BMC Struct Biol 2006, 6: 19.PubMed CentralView ArticlePubMedGoogle Scholar
- Laskowski RA: SURFNET: a program for visualizing molecular surfaces, cavities, and intermolecular interactions. J Mol Graph 1995, 13: 323–330. 307–308View ArticlePubMedGoogle Scholar
- Laurie ATR, Jackson RM: Q-SiteFinder: an energy-based method for the prediction of protein-ligand binding sites. Bioinforma Oxf Engl 2005, 21: 1908–1916.View ArticleGoogle Scholar
- Liang J, Edelsbrunner H, Woodward C: Anatomy of protein pockets and cavities: measurement of binding site geometry and implications for ligand design. Protein Sci Publ Protein Soc 1998, 7: 1884–1897.View ArticleGoogle Scholar
- Peters KP, Fauck J, Frömmel C: The automatic search for ligand binding sites in proteins of known three-dimensional structure using only geometric criteria. J Mol Biol 1996, 256: 201–213.View ArticlePubMedGoogle Scholar
- Brady GP Jr, Stouten PF: Fast prediction and visualization of protein binding pockets with PASS. J Comput Aided Mol Des 2000, 14: 383–401.View ArticlePubMedGoogle Scholar
- Li B, Turuvekere S, Agrawal M, La D, Ramani K, Kihara D: Characterization of local geometry of protein surfaces with the visibility criterion. Proteins 2008, 71: 670–683.View ArticlePubMedGoogle Scholar
- Hendlich M, Rippmann F, Barnickel G: LIGSITE: automatic and efficient detection of potential small molecule-binding sites in proteins. J Mol Graph Model 1997, 15: 359–363.View ArticlePubMedGoogle Scholar
- Levitt DG, Banaszak LJ: POCKET: a computer graphics method for identifying and displaying protein cavities and their surrounding amino acids. J Mol Graph 1992, 10: 229–234.View ArticlePubMedGoogle Scholar
- An J, Totrov M, Abagyan R: Pocketome via comprehensive identification and classification of ligand binding envelopes. Mol Cell Proteomics MCP 2005, 4: 752–761.View ArticlePubMedGoogle Scholar
- Coleman RG, Sharp KA: Travel depth, a new shape descriptor for macromolecules: application to ligand binding. J Mol Biol 2006, 362: 441–458.View ArticlePubMedGoogle Scholar
- Kleywegt GJ, Jones TA: Detection, delineation, measurement and display of cavities in macromolecular structures. Acta Crystallogr D Biol Crystallogr 1994, 50(Pt 2):178–185.View ArticlePubMedGoogle Scholar
- Ho CM, Marshall GR: Cavity search: an algorithm for the isolation and display of cavity-like binding regions. J Comput Aided Mol Des 1990, 4: 337–354.View ArticlePubMedGoogle Scholar
- Dennis S, Kortvelyesi T, Vajda S: Computational mapping identifies the binding sites of organic solvents on proteins. Proc Natl Acad Sci U S A 2002, 99: 4290–4295.PubMed CentralView ArticlePubMedGoogle Scholar
- Kortvelyesi T, Silberstein M, Dennis S, Vajda S: Improved mapping of protein binding sites. J Comput Aided Mol Des 2003, 17: 173–186.View ArticlePubMedGoogle Scholar
- Ruppert J, Welch W, Jain AN: Automatic identification and representation of protein binding sites for molecular docking. Protein Sci Publ Protein Soc 1997, 6: 524–533.View ArticleGoogle Scholar
- Verdonk ML, Cole JC, Watson P, Gillet V, Willett P: SuperStar: improved knowledge-based interaction fields for protein binding sites. J Mol Biol 2001, 307: 841–859.View ArticlePubMedGoogle Scholar
- Bliznyuk AA, Gready JE: Simple method for locating possible ligand binding sites on protein surfaces. J Comput Chem 1999, 20: 983–988.View ArticleGoogle Scholar
- Campbell SJ, Gold ND, Jackson RM, Westhead DR: Ligand binding: functional site location, similarity and docking. Curr Opin Struct Biol 2003, 13: 389–395.View ArticlePubMedGoogle Scholar
- Glick M, Robinson DD, Grant GH, Richards WG: Identification of ligand binding sites on proteins using a multi-scale approach. J Am Chem Soc 2002, 124: 2337–2344.View ArticlePubMedGoogle Scholar
- Sotriffer C, Klebe G: Identification and mapping of small-molecule binding sites in proteins: computational tools for structure-based drug design. Farm Soc Chim Ital 1989 2002, 57: 243–251.Google Scholar
- Andricopulo AD, Salum LB, Abraham DJ: Structure-based drug design strategies in medicinal chemistry. Curr Top Med Chem 2009, 9: 771–790.View ArticlePubMedGoogle Scholar
- Waszkowycz B, Clark DE, Gancia E: Outstanding challenges in protein-ligand docking and structure-based virtual screening. Wiley Interdiscip Rev Comput Mol Sci 2011, 1: 229–259.View ArticleGoogle Scholar
- Cavasotto CN, Orry AJW: Ligand docking and structure-based virtual screening in drug discovery. Curr Top Med Chem 2007, 7: 1006–1014.View ArticlePubMedGoogle Scholar
- Des Jarlais RL, Cummings MD, Gibbs AC: Virtual docking: how are we doing and how can we improve? Front Drug Des Discov Struct-Based Drug Des 21st Century 2007, 3: 81–103.Google Scholar
- Moitessier N, Englebienne P, Lee D, Lawandi J, Corbeil CR: Towards the development of universal, fast and highly accurate docking/scoring methods: a long way to go. Br J Pharmacol 2008, 153(Suppl 1):S7-S26.PubMed CentralPubMedGoogle Scholar
- Kontoyianni M, Madhav P, Suchanek E, Seibel W: Theoretical and practical considerations in virtual screening: a beaten field? Curr Med Chem 2008, 15: 107–116.View ArticlePubMedGoogle Scholar
- Tuccinardi T: Docking-based virtual screening: recent developments. Comb Chem High Throughput Screen 2009, 12: 303–314.View ArticlePubMedGoogle Scholar
- Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ: Basic local alignment search tool. J Mol Biol 1990, 215: 403–410.View ArticlePubMedGoogle Scholar
- Pearson WR: Rapid and sensitive sequence comparison with FASTP and FASTA. Methods Enzymol 1990, 183: 63–98.View ArticlePubMedGoogle Scholar
- Altschul SF, Madden TL, Schäffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ: Gapped BLAST and PSI-BLAST: a new generation of protein database search programs. Nucleic Acids Res 1997, 25: 3389–3402.PubMed CentralView ArticlePubMedGoogle Scholar
- Söding J, Biegert A, Lupas AN: The HHpred interactive server for protein homology detection and structure prediction. Nucleic Acids Res 2005, 33(Web Server issue):W244-W248.PubMed CentralView ArticlePubMedGoogle Scholar
- Zhang Y: Template-based modeling and free modeling by I-TASSER in CASP7. Proteins 2007, 69(Suppl 8):108–117.View ArticlePubMedGoogle Scholar
- Roy A, Kucukural A, Zhang Y: I-TASSER: a unified platform for automated protein structure and function prediction. Nat Protoc 2010, 5: 725–738.PubMed CentralView ArticlePubMedGoogle Scholar
- Liu TW, Tang G, Capriotti E: Comparative modeling: the state of the art and protein drug target structure prediction. Comb Chem High Throughput Screen 2011, 14: 532–547.View ArticlePubMedGoogle Scholar
- Nelson DL, Cox MM, Lehninger AL: Principles of Biochemistry. New York: Freeman; 2004.Google Scholar
- Murray RK: Harper’s Illustrated Biochemistry. New York: McGraw-Hill; 2003.Google Scholar
- Markley JL, Bax A, Arata Y, Hilbers CW, Kaptein R, Sykes BD, Wright PE, Wüthrich K: Recommendations for the presentation of NMR structures of proteins and nucleic acids (IUPAC Recommendations 1998). Pure Appl Chem 1998, 70: 117–142.View ArticleGoogle Scholar
- Atomic coordinate entry format version 3.3 http://www.wwpdb.org/documentation/format33/v3.3.html
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.