Nucleic Acids Research Advance Access originally published online on November 30, 2007
Nucleic Acids Research 2008 36(Database issue):D674-D678; doi:10.1093/nar/gkm911
Nucleic Acids Research, 2008, Vol. 36, Database issue D674-D678
© 2007 The Author(s)
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/2.0/uk/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Binding MOAD, a high-quality protein–ligand database
Mark L. Benson1,
Richard D. Smith2,
Nickolay A. Khazanov1,
Brandon Dimcheff3,
John Beaver3,
Peter Dresslar3,4,
Jason Nerothin4 and
Heather A. Carlson1,2,4,*
1Bioinformatics Graduate Program, 2Biophysics Research Division, University of Michigan, Ann Arbor, MI 48109, 3Torrey Path LLC, Ann Arbor, MI 48104 and 4Department of Medicinal Chemistry, University of Michigan, Ann Arbor, MI 48109, USA
*To whom correspondence should be addressed. Tel: +1 734 615 6841; Email: carlsonh{at}umich.edu
Received September 15, 2007. Revised October 5, 2007. Accepted October 5, 2007.
 |
ABSTRACT
|
|---|
Binding MOAD (Mother of All Databases) is a database of 9836
protein–ligand crystal structures. All biologically relevant
ligands are annotated, and experimental binding-affinity data
is reported when available. Binding MOAD has almost doubled
in size since it was originally introduced in 2004, demonstrating
steady growth with each annual update. Several technologies,
such as natural language processing, help drive this constant
expansion. Along with increasing data, Binding MOAD has improved
usability. The website now showcases a faster, more featured
viewer to examine the protein–ligand structures. Ligands
have additional chemical data, allowing for cheminformatics
mining. Lastly, logins are no longer necessary, and Binding
MOAD is freely available to all at
http://www.BindingMOAD.org.
|
INTRODUCTION
|
|---|
The field of structure-based drug design relies on high-quality
databases of protein–ligand structures to develop the
best computational tools. There are several available, including
but not limited to Binding MOAD (
1), PDBbind (
2), LPDB (
3),
Relibase (
4), BindingDB (
5), PDBLig (
6), MSDsite (
7), eF-Site
(
8), PDB-Ligand (
9), SuperLigands (
10), PLD (
11), HET-PDB (
12),
sc-PDB (
13), PDBsite (
14), Ligand Depot (
15), AffinDB (
16) and
K
iBank (
17). Each database has a unique focus and incorporates
different data content, chemical structures, and/or analysis
tools.
Our development of Binding MOAD focuses on providing the largest collection of high-quality, protein–ligand complexes. Each structure is hand curated by reading the crystallography paper which presents the structure in the literature; this is used to validate ligands and acquire binding affinities. Binding MOAD contains all appropriate protein–ligand complexes: protein–ligand, protein-cofactor and protein–ligand-cofactor. It also presents complexes even when no binding data is currently available. This makes it the largest database of its type. Here, we discuss the latest update to the Binding MOAD database, outlining improved access, the addition of new data and the incorporation of new tools.
|
BINDING MOAD COMPOSITION
|
|---|
Binding MOAD is constructed with a top-down approach, starting
with all entries in the PDB (
18) and eliminating structures
which are inappropriate. This is more efficient than a ground-up
approach of reading the literature as a whole to identify appropriate
complexes. Each entry in Binding MOAD must have resolution better
than 2.5 Å, and each entry must contain a valid ligand.
Valid ligands are biologically relevant small molecules and
can include agonists, antagonists, cofactors, inhibitors, allosteric
regulators, enzymatic products, etc. Covalently attached molecules
(covalent inhibitors or posttranslational modifications to the
protein) are not considered valid ligands. The focus is proteins
binding small molecules, so peptides larger than 10 amino acids
and chains of greater than four nucleic acids are not considered
valid ligands. Many small molecules present in a crystal structure
are not considered biologically relevant because they are part
of the crystallization matrix and an artifact of the protein
in an artificial condensed phase. Such molecules include solvents,
buffers, detergents and salts, but care must be taken because
some small molecules are valid ligands in some structures but
additives in others. Examples of such are sugars, membrane components,
small organic molecules (e.g. toluene) and metabolites (e.g.
citrate).
Figures 1 and
2 illustrate the wide distribution of
data available in Binding MOAD, in terms of binding affinity
and size, respectively.
View larger version (22K):
[in this window]
[in a new window]
[Download PowerPoint slide]
|
Figure 1. Distribution of binding affinities in Binding MOAD. Data is labeled as Kd (blue), IC50 (yellow) and Ki (red). For this figure, binding affinities were simply converted to free energies by RT x ln(affinity). While this conversion not strictly appropriate for Ki or IC50, it provides a comparison for the reader.
|
|
In Binding MOAD, proteins are grouped into families of 90% sequence
identity. By choosing one representative of each family (the
ligand with the best affinity), we can create a non-redundant
set which removes any skew resulting from proteins that are
heavily represented in the PDB (
18). Protein families are also
grouped by function using Enzyme Classification numbers and
our own non-enzymatic classifications. At the bottom of the
data page for each complex in Binding MOAD, the entire protein
family is reported and a link is provided to view all the data
for that functional class (
Figure 3). This allows a user to
start at a particular complex important to his/her research,
and from there, jump to other related structures.
|
GROWTH
|
|---|
Since its introduction in 2004, Binding MOAD has regularly expanded
its collection with new data. Originally it contained 5331 protein–ligand
complexes, and it has increased by almost 1500 each year, growing
to 6638 in 2005, 8250 in 2006 and 9836 with the latest update.
This steady growth mirrors the growth of the PDB; each year's
update has consistently shown that one-fourth of the PDB structures
meet our criteria for inclusion in Binding MOAD. Of the 9836
entries in the current version, 2950 (
30%) have binding data
curated from literature. It contains 3153 protein families and
5074 unique ligands.
As noted earlier, each crystallography paper is read to classify the ligands and extract affinity data for the ligand. Thus, adding new data to Binding MOAD involves reading a tremendous number of journal articles for manual annotation. After reading 10 000 papers, we have turned to automated methods to facilitate the process! A workflow tool, Binding Unstructured Data Analysis (BUDA), has been developed employing natural language processing (NLP) to evaluate and organize the papers for each update cycle. It identifies key sentences and phrases in papers and uses a weighted-scoring algorithm to rank the likelihood that the key sentences and phrases contain binding data. The NLP portion of BUDA is built upon the General Architecture for Text Engineering (GATE) framework (gate.ac.uk). The workflow portion of BUDA is used by the curators to organize the data for the annotation process. From the workflow interface, the curators can sort the articles by their weighted scores, review texts with highlights noting key phrases or sentences, and update the data into Binding MOAD. One of the key features is the ability to score a paper as lacking affinity data; it is a significant time-saving measure, rather than reading the paper in vain. While NLP can be used to speed up and guide the literature step, we unfortunately cannot use NLP to automatically extract the desired information. Many papers give affinity information for related systems when such information is unavailable for the exact complex in the crystal structure (e.g. affinities for wild-type protein are reported but the structures are all mutants or only the range of affinities is given for an entire inhibitor class). The data in Binding MOAD is for precisely the protein–ligand pair in the crystal structure, so data must be specific for that ligand bound to that exact protein. This evaluation must be made by hand.
|
AVAILABILITY
|
|---|
We have recently removed the need for users to login, and data
is now freely accessible to all private companies, non-profits
and foreign institutions. A comma-separated values (CSV) file
is available to obtain the binding data and ligand information,
organized by protein class and family. The CSV format was chosen
for wide portability. Structures are also available as biounit
files.
|
PLATFORM
|
|---|
Binding MOAD is built on proven technologies. The database is
based on the Java 2 Platform, Enterprise Edition (J2EE), using
an open-source JBoss Application Server, Enterprise JavaBeans
(EJB) and a MySQL database backend. These tools provide a standards-compliant,
easy-to-use website that unifies the presentation of structural,
chemical and binding data in one simple format. In particular,
the structure of the database allows us to expand the features
and add new data easily.
|
RECENT IMPROVEMENTS
|
|---|
A screenshot of the modified layout for a data page in Binding
MOAD is shown in
Figure 3. New data has been incorporated about
the valid ligands in each structure, including interactive 2D
pictures, chemical formulae and the corresponding SMILES strings.
As before, all ligands are noted as valid or invalid. When a
hetgroup is considered part of the protein (glycosylation, catalytic
metal, HEME group, etc.), it is not listed on the data page.
The greatest improvement comes as a new tool for viewing the complex in 3D. The GoCavViewer has been replaced by the EolasViewer; screenshot of the viewer is shown in Figure 4. As before, the viewer is capable of displaying the ligand pocket using both ball-stick and surface representations; the surfaces come from our code GoCAV which was specifically developed for Binding MOAD (19). However, EolasViewer incorporates significant improvements in the areas of performance, visual quality and back-end flexibility for future application development efforts.
View larger version (67K):
[in this window]
[in a new window]
[Download PowerPoint slide]
|
Figure 4. EolasViewer for 3ERK. The SB4 ligand is shown in ball in stick inside the pocket. The surfaces shown are the ligand surface in blue, the binding site in red and the solvent-exposed regions of the binding site are in green. (Top) The protein backbone is shown as a gray ribbon, and in the close-up (Bottom), the backbone is colored by B-factors.
|
|
The new viewer is built using the Eolus platform from Metamatics.
Like its predecessor, the Eolus-based viewer is built using
a Java framework, and the Binding MOAD website deploys it as
a WebStart application. Eolus uses Jogl (Java Bindings for OpenGL)
to fully utilize the 3D acceleration features available in nearly
all modern computers. These two technologies, Java WebStart
and OpenGL, provide nearly hands-free deployment of the software,
together with state-of-the-art performance and visual quality.
It takes advantage of rendering algorithms and OpenGL Shader
Language (GLSL) to provide enhanced representation styles. The
surface representation has been expanded to a fully transparent
polygon surface. The proteins are rendered as ribbons by default,
and the entire protein can now be rendered either as ribbon
or ball-stick (the GoCavViewer was limited to displaying only
residues that comprised the binding site). Finally, Eolus is
a platform for structural biology, being developed in conjunction
with this and other tools.
|
FUTURE DIRECTIONS
|
|---|
The data is currently organized with respect to protein structure
and function, but we will expand the organization of the ligands
by their chemical nature. At this time, ligands are annotated
by their 3-letter HET codes, but full names will soon be added.
A single-click search links all structures that contain the
same molecule, but that is the extent of cross-linking by ligand
data. We are adding similarity-based searches for the ligands.
This effort will incorporate the new remediated ligand data
released by the PDB consortium, and we look to cross-link our
information with other major databases that focus on proteins
and ligands. Furthermore, we look to use our text-mining tools
to extend our search for affinity data beyond the crystallography
literature. Lastly, Binding MOAD adds data once a year, but
we look to make this a semi-annual event, given the success
with NLP. Such NLP-based, text-mining approaches can be readily
applied to other bioinformatic projects. This technology can
be used to extract a wide variety of data—not just binding
information—from the huge body of literature available
today.
|
ACKNOWLEDGEMENTS
|
|---|
The authors wish to thank Steve Spronk for helpful discussion.
We thank ChemAxon (
www.chemaxom.com) for making web-based tools
accessible to non-profit websites. This work was originally
funded by a Young Investigator Award to H.A.C. from the Arnold
and Mabel Beckman Foundation. It is currently funded by a CAREER
Award to H.A.C. from the National Science Foundation (MCB 0546073).
R.D.S. thanks the Molecular Biophysics Training Program (funded
by the National Institutes of Health, GM 008270). N.A.K. thanks
the Training Program in Bioinformatics (funded by the National
Institutes of Health, GM 070449). Funding to pay the Open Access
publication charges for this article was provided by the HAC's
NSF grant (MCB 0546073).
Conflict of interest statement. None declared.
|
REFERENCES
|
|---|
- Hu L, Benson ML, Smith RD, Lerner MG, Carlson HA. Binding MOAD (Mother Of All Databases). Prot. Struct. Func. Bioinf. (2005) 60:333–340.[CrossRef]
- Wang R, Fang X, Lu Y, Yang CY, Wang S. The PDBbind Database: methodologies and updates. J. Med. Chem. (2005) 48:4111–4119.[CrossRef][Web of Science][Medline]
- Roche O, Kiyama R, Brooks CL. Ligand-protein database: linking protein-ligand complex structures to binding data. J. Med. Chem. (2001) 44:3592–3598.[CrossRef][Web of Science][Medline]
- Hendlich M, Bergner A, Gunther J, Klebe G. Relibase: design and development of a database for comprehensive analysis of protein-ligand interactions. J. Mol. Biol. (2003) 326:607–620.[CrossRef][Web of Science][Medline]
- Liu T, Lin Y, Wen X, Jorrisen RN, Gilson MK. BindingDB: a web-accessible database of experimentally determined protein-ligand binding affinities. Nucleic Acids Res. (2007) 35:D198–D201.[Abstract/Free Full Text]
- Chalk AJ, Worth CL, Overington JP, Chan AWE. PDBLIG: classification of small molecular protein binding in the Protein Data Bank. J. Med. Chem. (2004) 47:3807–3816.[CrossRef][Web of Science][Medline]
- Golovin A, Dimitropoulos D, Oldfield T, Rachedi A, Henrick K. MSDsite: a database search and retrieval system for the analysis and viewing of bound ligands and active sites. Prot. Struct. Func. Bioinf. (2005) 58:190–199.[CrossRef]
- Kinoshita K, Furui J, Nakamura H. Identification of protein functions from a molecular surface database, eF-site. J. Struct. Funct. Genomics (2002) 2:9–22.[CrossRef][Medline]
- Shin J-M, Cho D-H. PDB-ligand: a ligand database based on PDB for the automated and customized classification of ligand-binding structures. Nucleic Acids Res. (2005) 33:D238–D241.[Abstract/Free Full Text]
- Michalsky E, Dunkel M, Goede A, Preissner R. SuperLigands - a database of ligand structures derived from the Protein Data Bank. BMC Bioinformatics (2005) 6:122.[CrossRef][Medline]
- Puvanendrampillai D, Mitchell JBO. Protein Ligand Database (PLD): additional understanding of the nature and specificity of protein-ligand complexes. Bioinformatics (2003) 19:1856–1857.[Abstract/Free Full Text]
- Yamaguchi A, Iida K, Matsui N, Tomoda S, Yura K, Go M. Het-PDB Navi.: a database for protein-small molecule interactions. J. Biochem. (2004) 135:79–84.[Abstract/Free Full Text]
- Kellenberger E, Muller P, Schalon C, Bret G, Foata N, Rognan D. sc-PDB: an annotated database of druggable binding sites from the Protein Data Bank. J. Chem. Inf. Model. (2006) 46:717–727.[CrossRef][Web of Science][Medline]
- Ivanisenko VA, Pintus SS, Grigorovich DA, Kolchanov NA. PDBSite: a database of the 3D structure of protein functional sites. Nucleic Acids Res. (2005) 33:D183–D187.[Abstract/Free Full Text]
- Feng Z, Chen L, Maddula H, Akcan O, Oughtred R, Berman HM, Westbrook J. Ligand Depot: a data warehouse for ligands bound to macromolecules. Bioinformatics (2004) 20:2153–2155.[Abstract/Free Full Text]
- Block P, Sotriffer CA, Dramburg I, Klebe G. AffinDB: a freely accessible database of affinities for protein-ligand complexes from the PDB. Nucleic Acids Res. (2006) 34:D522–D526.[Abstract/Free Full Text]
- Zhang J, Aizawa M, Amari S, Iwasawa Y, Nakano T, Nakata K. Development of KiBank, a database supporting structure-based drug design. Comput. Biol. Chem. (2004) 28:401–407.[CrossRef][Web of Science][Medline]
- Berman HM, Westbrook J, Feng Z, Gilliland G, Bhat TN, Weissig H, Shindyalov IN, Bourne PE. The Protein Data Bank. Nucleic Acids Res. (2000) 28:235–242.[Abstract/Free Full Text]
- Smith RD, Hu L, Falkner JA, Benson ML, Nerothin JP, Carlson HA. Exploring protein-ligand recognition with Binding MOAD. J. Mol. Graph. Model. (2006) 24:414–425.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
|
|
 
R. A. Bauer, S. Gunther, D. Jansen, C. Heeger, P. F. Thaben, and R. Preissner
SuperSite: dictionary of metabolite and drug binding sites in proteins
Nucleic Acids Res.,
January 1, 2009;
37(suppl_1):
D195 - D200.
[Abstract]
[Full Text]
[PDF]
|
|
|
TOP
ABSTRACT
INTRODUCTION
