Scientific competency questions as the basis for semantically

Drug Discovery Today Volume 18, Numbers 17/18 September 2013
Scientific competency questions as the
basis for semantically enriched open
pharmacological space development
Kamal Azzaoui1, Edgar Jacoby14, Stefan Senger2, Emiliano Cuadrado Rodrı́guez3,
Mabel Loza3, Barbara Zdrazil4, Marta Pinto4, Antony J. Williams5, Victor de la Torre6,
Jordi Mestres7, Manuel Pastor7, Olivier Taboureau8, Matthias Rarey9,
Christine Chichester10, Steve Pettifer11, Niklas Blomberg12,a, Lee Harland13,
Bryn Williams-Jones13 and Gerhard F. Ecker4
Novartis Institutes for BioMedical Research, Novartis Pharma AG, Forum 1 Novartis Campus, CH-4056 Basel, Switzerland
GlaxoSmithKline, Medicines Research Centre, Stevenage SG1 2NY, UK
Grupo BioFarma-USEF, Departamento de Farmacologı́a, Facultad de Farmacia, Campus Universitario Sur s/n, 15782 Santiago de Compostela, Spain
University of Vienna, Department of Medicinal Chemistry, Pharmacoinformatics Research Group, Althanstrasse 14, 1090 Wien, Austria
Royal Society of Chemistry, 904 Tamaras Circle, Wake Forest, NC 27587, USA
Structural Computational Biology and National Bioinformatic Institute Unit, Structural Biology and Biocomputing Programme, Spanish National Cancer Research
Centre (CNIO), C/ Melchor Fernández Almagro 3, Madrid E-28029, Spain
Chemogenomics Laboratory, Research Programme on Biomedical Informatics, IMIM—Hospital del Mar Research Institute and Universitat Pompeu Fabra, Parc de
Recerca Biomèdica, Doctor Aiguader 88, 08003 Barcelona, Catalonia, Spain
Technical University of Denmark, Department of Systems Biology, Kemitorvet, Building 208, 2800 Lyngby, Denmark
Center for Bioinformatics, University of Hamburg, Bundesstraße 43, 20146 Hamburg, Germany
Swiss Institute of Bioinformatics, CALIPHO Group, CMU – Rue Michel-Servet 1, 1211 Geneva 4, Switzerland
School of Computer Science, The University of Manchester, Oxford Road, Manchester M13 9PL, UK
AstraZeneca R&D Mölndal, SE-431 83 Mölndal, Sweden
Connected Discovery Ltd, 27 Old Gloucester Street, London WC1N 3AX, UK
Janssen Research & Development, Turnhoutseweg 30, B-2340 Beerse, Belgium
Molecular information systems play an important part in modern data-driven drug discovery. They do
not only support decision making but also enable new discoveries via association and inference. In this
review, we outline the scientific requirements identified by the Innovative Medicines Initiative (IMI)
Open PHACTS consortium for the design of an open pharmacological space (OPS) information system.
The focus of this work is the integration of compound–target–pathway–disease/phenotype data for
public and industrial drug discovery research. Typical scientific competency questions provided by the
consortium members will be analyzed based on the underlying data concepts and associations needed to
answer the questions. Publicly available data sources used to target these questions as well as the need for
and potential of semantic web-based technology will be presented.
Drug discovery is a data-driven process [1]. The amount and
diversity of drug discovery data in the omics- and high-through-
Corresponding author:. Ecker, G.F. (
Present address: ELIXIR, EMBL-European Bioinformatics Institute, Wellcome
Trust Genome Campus, Hinxton, Cambridgeshire, CB10 1SD, UK.
put-driven paradigms has significantly grown to the point where
current relational data models have reached their performance
limits in terms of technical and scientific capabilities [2]. In addition to the need for data integration, it is recognized that providing capabilities for semantic inference is a key challenge and offers
a wealth of opportunities. Such a semantic molecular information
system was pioneered by the Wild group at Indiana University
1359-6446/06/$ - see front matter ß 2013 Elsevier Ltd. All rights reserved.
with the Chem2Bio2RDF system [3–6], which is based on the
Bio2RDF knowledge system provided earlier by Belleau at Laval
University, Montreal [7]. The Linking Open Drug Data (LODD)
project [8] is a comparable project within the World Wide Web
Consortium (W3C) [9] healthcare and life science interest group.
Recognizing the challenges and opportunities, the European
Union (EU) and European Federation of Pharmaceutical Industries
and Associations (EFPIA) decided to develop the Innovative Medicines Initiative (IMI) joint undertaking, and through this [10] the
Open Pharmacological Concepts Triple Store (Open PHACTS)
consortium [11]. The Open PHACTS project brings together academic and pharmaceutical partners to design and implement a
publicly available open pharmacological space. The project is
driven by scientific questions and use cases of various complexity
that apply to real-world drugs that, for the purpose of the Open
PHACTS project, we term ‘scientific competency questions’. By
focusing on standard use cases for drug discovery, the importance
of the competency questions is reflected in their general nature
rather than in the specific questions per se. The seemingly straightforward questions provide model scenarios that require careful
association (mapping) of multiple heterogeneous data across
diverse public domain databases. Core underlying data concepts
for this medicinal-chemistry-driven platform are ‘compound’,
‘target’, ‘pathway’ and ‘disease/phenotype’, all of them are relevant for the new fields of chemogenomics [12] and systems chemical biology [13,14].
As will become apparent from the analysis of the scientific
competency questions, the Open PHACTS system will build upon
the ideas of the Chem2Bio2RDF, Bio2RDF and LODD systems, to
address drug discovery research questions specifically. A key feature
of the Open PHACTS discovery platform is the openness for new
data additions that could include data from text mining of scientific
publications as well as opportunities for integration with proprietary or commercial data sources. Another key aspect is the development of novel visualization tools that facilitate the navigation
and knowledge extraction from all integrated data made available. It
is intended for the end-user tools not only to show how it is possible
to build relevant end-user applications on top of the Open PHACTS
platform but also to provide the bench scientist with immediate
value. Of course, all electronic data should be used with caution and
scientists need to be aware of its origin, provenance and reliability.
Thus, the goal in the Open PHACTS project is not simply to integrate
and query multiple databases but to provide a mechanism to understand how these results were obtained with attribution and provenance of individual data points [15].
A scientific competency question approach
When setting up an open, innovative, data integration and knowledge extraction platform, the first question arising is which out of
the more than 1000 open access databases [16] need to be integrated. This obviously depends on the type of queries the system
should allow the user to perform. Driven by the fact that in the first
instance the target audience will be bench scientists working in
drug discovery and development, the Open PHACTS consortium
develops a set of core use-cases to guide the project and to assist in
prioritizing the data sources selected for integration. Thus, the
original 22 Open PHACTS partners (eight EFPIA companies, 12
academic institutions, two SMEs; the project is continuously
Drug Discovery Today Volume 18, Numbers 17/18 September 2013
growing and currently comprises 28 partners now already) were
asked to provide ‘business’ questions that they believed would
enable progress in their specific research activities and drug discovery in general. Although most of the questions provided are not
challenging per se, answering them requires input from multiple
data sources hence needs in-depth knowledge of the data models
for a large set of systems. Thus, these represent a challenge to the
current information systems in use. Analysis of these questions
provides valuable information for the design of the graphical user
interface and guides the selection of data sources. In total 83
questions were collected in this approach, representing an effective survey of user needs and information priorities for preclinical
drug discovery research in pharmaceutical companies and academic institutions. The analysis of the questions followed a structured approach with input and critique from project partner
representatives. Because the results represent a clear and prioritized set of requirements and use cases for drug discovery research
projects, we believe they will have significant impact and use with
regard to knowledge management and systems design. The 83
questions were then grouped and prioritized using a point-based
voting system where each partner had one vote to rank the
importance of each question as high, medium or low. It is worth
noting that there were considerable differences in the rankings
between academic institutions from different domains (e.g. University of Vienna and Leiden University Medical Center), but
almost perfect correlation between EFPIA companies and academic institutions from the medicinal chemistry domain (e.g.
University of Vienna and AstraZeneca).
Prioritizing the 83 collected research questions and subsequent
analysis of the top 20 of these led to a deeper insight into the
actual information needs of researchers (in the pharma industry
as well as in academia and biotech) regarding data associations.
This analysis was carried out by extracting the key concepts
(compound, target, pathway, disease), as well as crucial mappings
between concepts implied in each question needed to start a
conventional data search. The application of this procedure
resulted in three main groups of 29 target/protein-related, 21
compound-related and 21 either disease- or pathway-related
questions. Minor groups from this analysis comprise gene/gene
family (six questions), substructure (five), protein family (three),
RNA (one) and assay (one).
To complete the above analysis and have an overview of what
requesters expect in highly ranked questions, we included for each
question the keywords compound, target, pathway and disease, in
each category of prioritization (Fig. 1). It immediately becomes
obvious that the concept ‘compound–target’ is predominantly
found in the highly prioritized questions. All questions contain
the concept ‘compound’, and 16 out of the top 20 questions also
refer to ‘target’. The top 20 questions were then grouped in two
clusters according to the complexity of information requested
about compound–target or compound–target–disease/pathway.
The questions and their clusters are summarized in Table 1, and
are herein analyzed in terms of required data concepts, required
data associations and potential public data sources needed.
Cluster I: compound–target
This first cluster contains 11 questions centered on basic pharmacology. These types of questions are usually asked in early drug
High priority
Medium priority
Low priority
Drug Discovery Today
The percentage of the 83 questions with respect to the key words:
compound, target, pathway, disease; categorized by priorities. It clearly
demonstrates that the concepts ‘compound’ and ‘target’ are the most
discovery phases at the hit- or lead-finding stages. Generally, the
user wishes to find out more about known interactions between a
compound or a set of compounds and a defined primary target
and/or other targets. The request can be expanded toward a target
family and/or the same target in different species. By defining an
activity threshold, the user expects a list of compounds that can be
useful for direct screening or to provide input for compound
library design targeting a new enzyme, for instance an oxidoreductase (Q1, Table 1).
Although looking deceptively simple, this question is challenging because it requires checking each compound for activity
against the >3000 oxidoreductases in both species concerned.
The question also illustrates the common approach of ‘target
families’ and hence highlights the need for a well maintained target
classification system. A manual search took two scientists three days
to retrieve the respective list of compounds. The prototype released
internally six months after the start of the project was able to
perform this query within a few seconds. Regarding information
on multiple targets or ADMET effects (Q2 and Q3, Table 1), questions are motivated by the need to understand the mechanism of
possible side-effects of drug candidates and drugs [17]. These are
typically asked when lead and drug candidates are assessed to decide
whether or not to progress them for further development. The
metabolism/toxicity-related issues could at least in part be answered
by providing predicted secondary pharmacology data for a given
compound of interest. The predicted data could be from an interrogation of existing bioactivity data or based on predictive in silico
models as applied, for example, in the Chemotargets software for
predicting the off-target pharmacology of small molecules [18], the
SEA-approach [19] or the PASS algorithm [20].
In some cases the user might be more interested in a profile of
activity rather than a single activity or interested in similar compounds with a similar activity profile (Q4 and Q9, Table 1). To
answer this type of question one has to provide a defined bioactivity interaction profile, and then search for compounds that
share similarity in terms of their bioactivity profile. This use case
addresses a typical lead-finding strategy searching for compounds
with different chemotypes but similar bioactivity profiles that are
also expected to share activity on new targets [21]. In comparison
with previous questions the complexity increases remarkably
when the definition of the query requires substructure matching
capabilities and similarity searching (Q5 and Q10, Table 1). To
query and answer the specific questions above might well go
beyond the capabilities of a simple, easy-to-use graphical user
interface (GUI). Thus, a set of end-user applications, so called
example applications (eApps), are developed within the consortium on top of a robust Open PHACTS services application programming interface (API) and will be discussed later. These eApps
are proof-of-concept studies to demonstrate the capability of the
Open PHACTS discovery platform and API to enable effective
services built on top of it. They comprise tools with advanced
querying capabilities as well as scientific applications addressing
specialized needs.
To answer questions such as Q6 and Q8 (Table 1), a certain level
of granularity in the gene classification systems is needed. In fact,
the answer to Q6 returns a list of chemical compounds with
structures that are active against protein kinase C (PKC)a or all
other members of the PKC subfamily of kinases. The question is a
typical homology-based hit-finding strategy applicable for projects
where a large knowledge base exists [22]. The answer to Q8, where
specific interactions are targeted such protein–protein interactions
(PPIs), needs a specific database for such request. Finally, an answer
to Q11 requires late-phase development clinical data. The question, motivated by the desire to assess clinical compounds, can be
part of both clusters because the clinical data can be linked to a
disease study. For the specified list of clinical compounds, the
system should provide the available clinical data in addition to the
bioactivity data. However, although incredibly useful, this is
beyond the reach of the Open PHACTS project in its current
definition. Nevertheless, the integration capabilities offered by
the Open PHACTS discovery platform definitely will allow extensions toward translational data.
Cluster II: compound–target–disease/pathway
The second cluster contains nine questions that deal with the
previous concepts of compound–target relationships but, in addition, information about more-complex pharmacology in context
of pathways and diseases is needed. The information requested in
most of the cases needs references associated with it (patents,
journal articles). These types of questions are usually asked during
the lead optimization phase or proof-of-concept studies. The first
question in the cluster (Q12, Table 1) is motivated by the fact that
patents constitute an indispensable information source for chemical compounds and biological targets. The complexity of the
question is twofold: first, all patents have to be retrieved that relate
to the compound and the disease of interest; in a second step, the
targets need to be extracted from the patent claims. However, the
retrieval and/or tagging of text [23], as well as the recognition of
targets within patents [24], is extremely challenging and subject to
active research.
Question 13 (Table 1) is an aggregate of different questions seen
already; owing to its composite nature the complexity is extremely
high. First, as in Q3, the bioactivity of a compound for a specific
target needs to be established. As in Q11, the compound is in
preclinical or clinical phase and is an advanced compound
that should have public data available. The link to the relevant
% of the questions containing
the key words
Drug Discovery Today Volume 18, Numbers 17/18 September 2013
Drug Discovery Today Volume 18, Numbers 17/18 September 2013
The top 20 research questions
Question number
Cluster I
Cluster II
Give me all oxidoreductase inhibitors active <100 nM in human and mouse
Given compound X, what is its predicted secondary pharmacology? What are the on- and off-target safety concerns for a
compound? What is the evidence and how reliable is that evidence (journal impact factor, KOL) for findings associated with
a compound?
Given a target, find me all actives against that target. Find/predict polypharmacology of actives. Determine ADMET profile
of actives
For a given interaction profile – give me similar compounds
The current Factor Xa lead series is characterized by substructure X. Retrieve all bioactivity data in serine protease assays
for molecules that contain substructure X
A project is considering protein kinase C alpha (PRKCA) as a target. What are all the compounds known to modulate the
target directly? What are the compounds that could modulate the target directly? I.e. return all compounds active in assays
where the resolution is at least at the level of the target family (i.e. PKC) from structured assay databases and the literature
Give me all active compounds on a given target with the relevant assay data
Identify all known protein–protein interaction inhibitors
For a given compound, give me the interaction profile with targets
For a given compound, summarize all ‘similar compounds’ and their activities
Retrieve all experimental and clinical data for a given list of compounds defined by their chemical structure (with options
to match stereochemistry or not)
For my given compound, which targets have been patented in the context of Alzheimer’s disease?
Which ligands have been described for a particular target associated with transthyretin-related amyloidosis, what is their
affinity for that target and how far are they advanced into preclinical/clinical phases, with links to publications/patents
describing these interactions?
Target druggability: compounds directed against target X have been tested in which indications? Which new targets have
appeared recently in the patent literature for a disease? Has the target been screened against in AZ before? What
information on in vitro or in vivo screens has already been performed on a compound?
Which chemical series have been shown to be active against target X? Which new targets have been associated with
disease Y? Which companies are working on target X or disease Y?
Which compounds are known to be activators of targets that relate to Parkinson’s disease or Alzheimer’s disease
For my specific target, which active compounds have been reported in the literature? What is also known about upstream
and downstream targets?
Compounds that agonize targets in pathway X assayed in only functional assays with a potency <1 mM
Give me the compound(s) that hit most specifically the multiple targets in a given pathway (disease)
For a given disease/indication, give me all targets in the pathway and all active compounds hitting them
publications and patents renders it similar to Q12 and Q17. The
compound–publication and target–publication associations are
needed. In some questions such as Q14 there is a need to compare
the resulting data with proprietary in-house data, which poses a
technical challenge because the publicly available Open PHACTS
discovery platform needs to be able to integrate proprietary data.
This imposes another level of complexity because it requires secure
access. In addition, the whole issue of licensing needs to be
addressed, which is extremely demanding when it comes to mixing public and private data into one platform.
Question 15 can be related to Q5 and Q10. However, what is
new in regard to the related questions is the specification of the
concept of a chemical series; hit compounds would need to be
clustered around the parent series scaffold. Although various
computational definitions for this task exist, there still remains
the challenge of agreeing on the method to be used in the first
instance. Open PHACTS is thus also actively working on providing
standards agreed and widely adopted by the community, also
including the pharmaceutical industry. Finally, the question of
new targets associated with a disease and the competitive landscape around a target or disease are factors interrogated in Q15.
Questions 17 and 18 are similar to Q3. More specifically, they
require the knowledge of the pathway(s) where the specific target
is representing a node. The capability to extract the upstream and
downstream interaction partners from the pathway map adds an
extra degree of complexity. Answering Q19 and Q20 imposes two
levels of complexity. First, the compounds that hit the known
targets in a given canonical disease pathway would need to be
identified. In a second step, the most-specific compounds would
need to be extracted. Both questions are relevant in the context of
target and lead identification in the newer biology-driven drug
discovery paradigm [25,26].
Data association and/or data sources
The analysis of the scientific competency questions demonstrates
that the data and associations between the concepts ‘chemical
compound’, ‘molecular target’, ‘biological pathway’ and ‘disease’
are essential to address the questions. One can visualize the complexity of the data associations in a network fashion (Fig. 2). With
this network, we intend to summarize the data and associations
needed to answer just the top 20 prioritized questions and thus to
lay the foundation for selecting public databases that provide the
respective information. In this section we will go through the
details of the data needed and the public databases that provide
such data (details and references about the data sources are listed
in Table 2).
Drug Discovery Today Volume 18, Numbers 17/18 September 2013
Bioactivity data
Clinical data
Chemical info
Gene classification
Side effects
Drug Discovery Today
Network of data associations needed to answer the top-ranked scientific competency questions. The network reflects a cartoon that summarizes the data
associations that are needed to target the top 20 research questions.
Chemistry node
Biology nodes
The information about drug names and the required search mode
depends strongly on the use of correct chemical names such as
systematic names, trade names or synonyms. Such dictionaries of
chemical name–structure associations are available from a number
of chemical databases including ChEBI, DrugBank and ChemSpider [27]. In addition, chemical structure information is needed for
defining queries such as substructure matching or similarity
searches. This point stresses the need for data quality in general
and especially on the exactness of representation of chemical
structures, including stereochemistry, tautomers and protonation
states. The quality of data in public domain databases has been
discussed in a number of publications and highlights the need for a
consistent and publicly described framework for normalization
[28,29]. Within the Open PHACTS project those chemical-related
queries will be provided via an interface to the ChemSpider
molecular information system. For the compound–bioactivity
association several highly popular databases will be integrated
in the Open PHACTS discovery platform. In the first instance,
those are ChEMBL, Chebi and DrugBank, which comprise largescale public sources for the compound and bioactivity data. The
bioactivity databases provide data at the level of primary activity at
a single concentration of compound or data from dose–responsebased experiments such as IC50 and/or EC50, Ki or Kb. Hence, there
is a need for the system to handle quantitative data within the
semantic interoperability framework.
It is important to recognize that a target in a given assay can be of
heterogeneous nature, including proteins, cells or even whole
organisms. To illustrate this, a search performed in ChEMBL for
propafenone, a class 1C antiarrhythmic agent, revealed target
name instances such as Ratus norwegicus, Plasmodium falciparum,
CYP 450 2D6 and CCRF CEM 1000. It is thus necessary to represent
the target–protein–gene associations in full detail. The gene–phenotype association can be found in the OMIM or GO data systems.
More specifically, gene–side-effect associations can be found in
specific literature on safety profiling [17]. The IUPHAR database
provides structured pharmacological data on ion channels, Gprotein-coupled receptors (GPCRs) and nuclear receptors. The
SIDER database at EMBL contains information on marketed medicines and their recorded adverse drug reactions. However, it
should be mentioned that the definition of secondary pharmacological data is ‘fuzzy’ given that a compound can have more than
one primary target in the view of its polypharmacology, which
finally defines its pharmacodynamic in vivo profile [30].
Protein–pathway association data are provided by the WikiPathways and Reactome data repositories. Also the GO classification of
genes will be useful in this context. Gene–disease association data
are part of the OMIM and Diseaseome datasets. This nicely illustrates the need for granularity in the gene classification systems.
Because the previously explored target families such target ontologies have been elaborated [31]. The key difficulty is to have
Drug Discovery Today Volume 18, Numbers 17/18 September 2013
Summary view of data sources, content and data concepts that are of interest for an open pharmacological space, recorded in March
2012. Emphasized is the open and free public access of the data sources
Database name
Internet resource URL/source
Specification of data and information available
ADME-AP provides comprehensive information about all classes of ADME-associated
proteins described in the literature including physiological function of each protein,
pharmacokinetic effect, ADME classification, direction and driving force of disposition,
location and tissue distribution, substrates, synonyms, gene name and protein
availability in other species. Cross-links to other databases are also provided to facilitate
the access of information about the sequence, 3D structure, function, polymorphisms,
genetic disorders, nomenclature, ligand binding properties and related literatures of
each protein. ADME-AP currently contains entries for 321 proteins and 964 substrates
Cancer Central Clinical
Database (C3D)
C3D is a clinical trials data management system that collects clinical trial data using
standard case report forms based on common data elements. It utilizes security
procedures to protect patient confidentiality and maintain an audit trail as required by
FDA regulations
ChEBI (Chemical Entities of Biological Interest) is a freely available dictionary of chemical
compounds, with IUPAC and NC-IUBMB endorsed terminology. Currently three data
sources have been incorporated into ChEBI, namely KEGG Ligand, IntEnz and Chemical
ChemBank is a freely available chemoinformatics database. The data are derived from
small molecules and small-molecule screens and resources for studying these data. It
was developed through a collaboration with the Chemical Biology Program and Platform
at the Broad Institute of Harvard and MIT
ChEMBL is a database of bioactive drug-like small molecules. This database also contains
2D structures, calculated properties (e.g. log P, molecular weight, Lipinski parameters)
and abstracted bioactivities (e.g. binding constants, pharmacology and ADMET data)
ChemSpider is a free chemical structure database from the Royal Society of Chemistry
providing fast text and structure/substructure search access to over 26 million structures
from over 400 data sources is a registry and results database of federally and privately supported
clinical trials conducted in the USA and around the world. It gives information about a
trial’s purpose, who may participate, locations and phone numbers, among others
Diseases Database
The Diseases Database is a database that underlies a free website that provides
information about the relationships between medical conditions, symptoms and
DISEASOME provides the genes that are associated with diseases and potentially
deleterious SNPs among the genes that are strongly associated with specific diseases
and clinical phenotypes. Currently, it contains 14,674 records on genetic variation and
109,715 records on genes related to human diseases
The DrugBank database combines detailed drug (i.e. chemical, pharmacological and
pharmaceutical) data with comprehensive drug target (i.e. sequence, structure and
pathway) information. The database (version 3.0) contains 6708 drug entries. 4229
nonredundant protein (i.e. drug target/enzyme/transporter/carrier) sequences are linked
to these drug entries
GenBank is the NIH genetic sequence database, an annotated collection of all publicly
available DNA
GO Database
The GO (Gene Ontology) database is a relational database comprising the GO ontologies
and the annotations of genes and gene products to terms in the GO. The advantage of
housing the ontologies and annotations in a single database is that powerful queries can
be performed over annotations using the ontology
The Human Metabolome Database (HMDB) is a freely available electronic database
containing detailed information about small molecule metabolites found in the human
body. The database (version 2.5) contains over 7900 metabolite entries. Additionally,
approximately 7200 protein (and DNA) sequences are linked to these metabolite entries
IntAct provides a freely available, open source database system and analysis tools for
protein interaction data. All interactions are derived from literature curation or direct
user submissions and are freely available
InterPro is an integrated database of predictive protein signatures used for the
classification and automatic annotation of proteins and genomes. It classifies sequences
at superfamily, family and subfamily levels, predicting the occurrence of functional
domains, repeats and important sites. InterPro adds in-depth annotation, including GO
terms, to the protein signatures
Drug Discovery Today Volume 18, Numbers 17/18 September 2013
Internet resource URL/source
Specification of data and information available
IUPHAR Database
The IUPHAR Database is the official database of the IUPHAR Committee on Receptor
Nomenclature and Drug Classification. It incorporates detailed pharmacological,
functional and pathophysiological information on G-protein-coupled receptors, voltagegated ion channels, ligand-gated ion channels and nuclear hormone receptors
Kinetic data of biomolecular interaction (KDBI) is a collection of experimentally
determined kinetic data of protein–protein, protein–RNA, protein–DNA, protein–ligand,
RNA–ligand, DNA–ligand-binding or reaction events described in the literature
MetaCyc is a database of nonredundant, experimentally elucidated metabolic pathways.
MetaCyc contains more than 1100 pathways from more than 1500 different organisms.
MetaCyc is curated from the scientific experimental literature and contains pathways
involved in primary and secondary metabolism, as well as associated compounds,
enzymes and genes
OMIM (online Mendelian inheritance in man) is a comprehensive, authoritative and
timely compendium of human genes and genetic phenotypes. The full-text, referenced
overviews in OMIM contain information on all known Mendelian disorders and over
12,000 genes. OMIM focuses on the relationship between phenotype and genotype
The 2P2I database stores structural information about PPIs with known inhibitors and
provides a useful tool for biologists to assess the potential druggability of their interfaces
PDSP (Psychoactive Drug Screening Program) provides screening of novel psychoactive
compounds for pharmacological and functional activity at cloned human or rodent CNS
receptors, channels and transporters
The PharmGKB database is a central repository for genetic, genomic, molecular and
cellular phenotype data and clinical information about people who have participated in
pharmacogenomics research studies. The data includes, but is not limited to, clinical and
basic pharmacokinetic and pharmacogenomic research in the cardiovascular,
pulmonary, cancer, pathways, metabolic and transporter domains
Protein Data Bank (PDB)
The PDB archive contains information about experimentally determined structures of
proteins, nucleic acids and complex assemblies. Users can perform simple and advanced
searches based on annotations relating to sequence, structure and function
PubChem is a free database of small molecules and information on their biological
activities. The system is maintained by the National Center for Biotechnology
Information (NCBI), a component of the National Library of Medicine, which is part of the
United States National Institutes of Health (NIH). It is linked to NIH PubMed/Entrez
REACTOME is an open source, open access, manually curated knowledgebase of
biological pathways in humans. Pathway annotations are authored by expert biologists,
in collaboration with Reactome editorial staff and cross-referenced to many
bioinformatics databases
SIDER (Side Effect Resource) contains information on marketed medicines and their
recorded adverse drug reactions. The information is extracted from public documents
and package inserts. The available information includes side-effect frequency, drug and
side-effect classifications as well as links to further information, for example drug–target
STITCH (search tool for interactions of chemicals) is a searchable database that integrates
information about interactions from metabolic pathways, crystal structures, binding
experiments and drug–target relationships. This database contains interactions for
between 300,000 small molecules and 2.6 million proteins from 1133 organisms
SuperTarget is a database that contains more than 2500 target proteins, which are
annotated with about 7300 relationships to 1500 drugs; the vast majority of entries have
pointers to the respective literature source
TOXLINE records provide bibliographic information covering the biochemical,
pharmacological, physiological and toxicological effects of drugs and other chemicals. It
contains over 4 million bibliographic citations, most with abstracts and/or indexing
terms and CAS registry numbers
UniProt is a freely accessible database of protein sequence and functional information,
many of it derived from genome sequencing projects. It contains a large amount of
information about the biological function of proteins derived from the research literature
WikiPathways is an open, collaborative platform dedicated to the curation of biological
pathways. It was built on the MediaWiki software and thus enables a broad usage by the
entire community
TABLE 2 (Continued )
Database name
ontologies at the genome-wide level, with the need for a sustained
development of ontologies for medicinal chemistry targets. This
further expands toward ontologies especially dealing with ADME
and toxicity. Because of its intrinsic complexity a toxicology
ontology represents a particular challenge and is addressed within
the IMI eTox Consortium [32]. Finally, to retrieve assay data not
only indirectly from databases but also directly from the primary
literature poses an immense challenge for chemical and biological
text mining [33–35]. Some specific questions were related to PPIs,
which again stresses the need for granularity in the gene classification systems. With the 2P2I database a specific data source regarding PPI inhibitors is emerging. The value of PPI datasets such as
IntAct would need to be evaluated in the perspective of its relevance to contribute to the answer for Q8. One of the highestprioritized questions was related to oxidoreductase enzymes. In
this case, the Enzyme Commission (EC) classification system in
UniProt provides a source for the enzyme classification.
Patent and publication node
Patents constitute a valuable source of information for chemical
compounds and biological targets. To answer some of the questions related to patents, all patents that relate to the compound
and the disease of interest have to be retrieved. In a second step,
the targets need to be extracted from the patent claims. Currently,
no public domain database with the required specific data associations exists. The European Bioinformatics Institute (EBI) in collaboration with the European Patent Office (EPO) is working toward
this goal [36]. Also, the SureChem family of products [37] offers
access to a comprehensive international patent collection, which
is normalized and curated. Finally, the IBM patent analytics platform is a pharmaceutical industry consortium focusing on the
same area [38,39]. However, accurate extraction of chemical structure information from patent documents constitutes a key challenge [40], which needs considerable attention. With respect to
publications, the most popular public database used to access
references of publications is PubMed. Although target and chemical names can be directly extracted from abstracts, extracting all
chemicals in a publication with its associated data will certainly
remain a significant challenge.
Finally, it has to be pointed out that the main goal of Open
PHACTS – connecting publicly available databases – inherently
puts a ‘data bias’ onto the whole system. In addition, the selection
of data sources is also heavily influenced by their respective
license. Thus, although some data sources can be better than
others, the final choice is based on a list of criteria rather than
on best coverage of the respective domain. These include, among
others, the compatibility of the license with the Open PHACTS
license, the availability of an RDF version, regular updates and
maintenance, and data quality.
OPS example end-user applications (eApps)
The Open PHACTS infrastructure will consist of a series of software
components that together form a platform that multiple applications can be built upon. All applications access the underlying data
via an API that provides access to optimized queries of the system.
Form-based queries within the core Open PHACTS interface (Open
PHACTS explorer) will enable the bench scientist to address key
use cases like the retrieval of compound, target and pathway
Drug Discovery Today Volume 18, Numbers 17/18 September 2013
information. In addition to the API, several other end-user applications will be co-developed with the Open PHACTS discovery
platform, such as a target dossier, a polypharmacology browser, a
chembio navigator and an application specialized for linking to
the toxicity data store established under the framework of the eTox
project. The relevance of these tools is not only to show how it is
possible to build relevant end-user applications on top of the core
platform but also to provide the bench scientist with immediate
value. The target dossier is designed to provide a comprehensive
view on pharmacologically relevant targets to answer questions
regarding druggability, tissue expression profiles and implications
in pathways, disease states and physiological mechanisms. The
polypharmacology browser is a tool that aims at enabling scientists
to define a target profile of interest and interrogate the Open
PHACTS discovery platform for compounds having affinity for
some or all of those targets, as well as for additional proteins that
might show a degree of cross-pharmacology with any of the targets
in the profile. The chembio navigator will enable intuitive browsing of the chemical and biological spaces. With filtering by various
structure and physicochemical descriptors, as well as by chemical
substructure and bioactivity data, the chembio navigator enables
an interactive analysis of datasets. Connected to the Open
PHACTS discovery platform, it will enable the user to drill down
into the primary data via hyperlinks.
Heterogeneity of data and chemical data integration
When integrating databases of different origin and built on different concepts (compound, target, pathway, disease, among
others), the heterogeneous nature of the data needs to be emphasized, as does the diverse nature of the quality of data contained in
the various databases. This poses a significant challenge to their
integration within the framework of more-classical relational
databases, especially for compound databases where the mappings
between chemical compounds are generally based on an electronic
structure format such as molfiles, SMILES and InChIs [41]. Indeed,
currently there is no public information system available that is
based on the relational database model and that allows the type of
more complex competency questions posed here to be addressed.
The majority of existing information systems has succeeded in
integrating associations between pairs of the above mentioned
data concepts. The ChEMBL system for instance provides an
excellent integration of the compound–target concepts including
an ontology for the represented targets. Also, within pharmaceutical companies, large-scale integration of chemical structure and
bioactivity data – the SAR estate – has been successfully completed.
AstraZeneca, for instance, reported recently the development of a
relational enterprise application containing 45 million unique
chemical structures from 18 internal and external sources [42].
The system merges compound-to-assay-to-result-to-target relationships enabling users to search with drug names, synonyms,
chemical structures, patent numbers and target protein identifiers,
at a scale not previously available. Similarly, the ChemSpider
database has mapped together 26 million chemical structures from
over 400 distinct data sources, mostly available online, and provides links out to PubChem (where assay information can be
accessed), to Google Scholar and Pubmed (where articles can be
accessed) and to Google Patents and the SureChem patent service
where patents can be retrieved. ChemSpider also has links out to
the ligands on Protein Data Bank, and to the majority of the
compound-based databases mentioned in this article including
ChEBI, ChEMBL, DrugBank and HMDB. The handling of complex
chemical queries such as structure-, substructure- and similaritybased searching for the Open PHACTS discovery platform will be
handled by accessing web services provided by the ChemSpider
database. ChemSpider holds the role of compound-based data
aggregator for the project and brings together compound-based
datasets, providing the appropriate database mappings to the
triple store, and uses stringent controls for data processing and
validation to ensure as high a quality as possible at the compound
level. Crowdsourcing capabilities are available on the platform so
that new compound data can be deposited and existing data can be
annotated and curated. ChemSpider contributes a subset of the
entire dataset available to the core platform. The data slice coincides with the data sources identified to be of value to the Open
PHACTS project and includes ChEBI, ChEMBL and DrugBank,
together with ChemSpider identifier mappings, chemical names
and synonyms and structure representations including SMILEs
and InChIs. On the basis of feedback from the EFPIA members
of the Open PHACTS consortium, and in alignment with chemical
compound standardization recommendations from the FDA [43],
the ChemSpider database will be re-standardized in the near future
to ensure compatibility between FDA standards and the Open
PHACTS discovery platform. However, although all these successful efforts demonstrate that it is indeed possible to bring together
the majority of the relevant chemical space, significant issues
remain, such as synchronized updates and maintaining integrity
of links, quality control of the chemical structures and, last but not
least, the big issue of normalization and standardization of chemical structures. The latter comprises protonation states, salts and
Further databases and need for ontologies
The focus aim of the Open PHACTS project is to cover the four
high-priority data concepts: chemical compound, molecular target, biological pathway and disease, including the relevant ontologies. Via mappings of identifiers, the semantic approach opens a
model-free approach for integration [44]. The data model is
included in the data and, in principle, no further specification
is needed. An explicit data model, as implied in relational databases, restricts the query capabilities. Thus, Open PHACTS fosters
the reutilization and integration of the large public investment in
data sources rather than generation of additional databases.
Although the concepts ‘chemical compound’, ‘biological target’
and ‘biological pathway’ reference databases are well established,
relatively few data sources can be found around the concepts
‘disease/phenotype’. The development of public mechanistically
determined clinical trial data sources are thus further encouraged
and will help to advance application in translational medicine
[45]. In addition, several other IMI projects, such as eTRIX,
DDMORE and EHR4CR, are focusing on this area. A typical application would be, for instance, the systematic repurposing of
marketed and experimental therapeutics by enabling the discovery of new potential therapeutic indications [46,47]. In early drug
discovery, the development of databases of microscopic cell and
organism phenotype data as generated in high-content screening
(HCS) will potentially enable further closing of the gap between
molecular and physiological observations. A large number of
phenotypic compound screens are ‘black box’ screens where the
identification of active compounds needs to be followed-up by in
silico and in vitro chemogenomics target fishing approaches to
elucidate their possible mechanisms of action. Although there is
no concrete scientific competency question for this among the top
20 questions, many of them would be part of an HCS work-up
workflow. The analysis of the scientific competency questions also
demonstrates the need for further development of ontologies.
Ontologies are computable formal explicit specifications of a
shared concept of a knowledge domain. They define entities
and the relationships between these entities and express this
knowledge in a formal and computable way. These ontologies
and classifications are indeed essential to allow not only data
aggregation and navigation but also inference-based modeling.
In the perspective of medicinal-chemistry-driven applications
as outlined in the scientific competency questions, comprehensive
target and ADME ontologies are of the utmost importance, as is the
contribution of ADME data to the public domain [48]. The GO
classifications enable interesting aggregations for chemical biology applications such as providing known compounds active
through a given biological mechanism like apoptosis or autophagy. The InterPro classification enables the aggregation of data
for a given protein domain type. The ChEBI ontology [49,50]
provides a chemical information ontology that enables aggregation of compound classes like lipids (or many others), and was
established to provide biologists with a more intuitive access to the
chemical space. Finally, as outlined above, ChemSpider allows
large-scale chemistry integration and also aims at providing highquality chemical compound standardization.
Concluding remarks
The top 20 priority questions presented herein are prevalently
motivated by chemogenomics and chemical biology applications
that are essential to the early drug discovery process. They refer to
the basic concepts of compound–target and compound–target–
disease/pathway relationships. Other questions are motivated by
the more general scientific need to enable integrative assessment
of information on compounds, targets, pathways and diseases. The
first examples of such capabilities were provided by the Wild group
using their semantic system to investigate the genetic basis of sideeffects of thiazolinedione drugs, resulting in a hypothesis for the
recently discovered cardiac side-effects of rosiglitazone (Avandia1)
and a prediction for pioglitazone, which is backed up by recent
clinical studies [51]. These scientific motivations are also well
aligned with the overall goal of the Open PHACTS project to
deliver an open, integrated and sustainable chemistry, biology
and pharmacology resource for drug discovery. The openness of
the system for profit-based and non-profit-based organizations, as
well as the long-term sustainability plan for the Open PHACTS
discovery platform and API, will be crucial to its success in industrial and academic drug discovery.
The functional Open PHACTS discovery platform will enable
the retrieval of existing data and associations. Given the complexity of biological data, critical analysis by the end-user scientist will
be needed to interpret the findings appropriately. This critical
assessment will be especially important for the inferred knowledge
Drug Discovery Today Volume 18, Numbers 17/18 September 2013
that is enabled by the semantic rule-set and inference technologies. In addition, with the enormous progress in modern biology,
computational scientists are confronted with the need for flexible
integration of a wide source of data, which requires new
approaches. Finally, the concepts and technological solutions
outlined and pursued for the open pharmacological space could
be easily expanded to create an ecosystem of interoperable open
spaces [open transcriptomic space (OTS), open genomic space
(OGS): generally OXS] in the life science area.
Drug Discovery Today Volume 18, Numbers 17/18 September 2013
The authors should like to thank all colleagues participating in the
Open PHACTS consortium for discussions and insights leading to
this consolidated view. The research leading to these results has
received support from the Innovative Medicines Initiative Joint
Undertaking under grant agreement no. [115191], resources of
which are composed of financial contribution from the European
Union’s Seventh Framework Programme (FP7/2007-2013) and inkind contribution of EFPIA companies.
1 Barnes, M.R. et al. (2009) Lowering industry firewalls: pre-competitive informatics
initiatives in drug discovery. Nat. Rev. Drug Discov. 8, 701–708
2 Mons, B. et al. (2011) The value of data. Nat. Genet. 43, 281–283
3 Chen, B. et al. (2010) Chem2Bio2RDF: a semantic framework for linking and data
mining chemogenomic and systems chemical biology data. BMC Bioinformatics 11,
4 Zhu, Q. et al. (2010) WENDI: a tool for finding non-obvious relationships between
compounds and biological properties, genes, diseases and scholarly publications. J.
Cheminform. 2, 6
5 Zhu, Q. et al. (2011) Semantic inference using chemogenomics data for drug
discovery. BMC Bioinformatics 12, 256
6 Wild, D.J. et al. (2011) Systems chemical biology and the Semantic Web: what they
mean for the future of drug discovery research. Drug Discov. Today 17, 469–474
7 Belleau, F. (2008) Bio2RDF: towards a mashup to build bioinformatics knowledge
systems. J. Biomed. Inform. 41, 706–716
8 Samwald, M. (2011) Linked open drug data for pharmaceutical research and
development. J. Cheminform. 3, 19
9 W3C. Available at: (accessed May 2013)
10 Innovative Medicines Initiative. Available at: (accessed
May 2013)
11 Open PHACTS. Available at: (accessed May 2013)
12 Jacoby, E. (2011) Computational chemogenomics. WIREs Comput. Mol. Sci. 1, 57–67
13 Oprea, T.I. et al. (2011) Computational systems chemical biology. Methods Mol. Biol.
672, 459–488
14 Oprea, T.I. et al. (2007) Systems chemical biology. Nat. Chem. Biol. 3, 447–450
15 Bechhofer, S. et al. (2010) Why linked data is not enough for scientists. Sixth IEEE eScience Conference Available at: In:
16 Galperin, M.Y. and Fernandez-Suarez, X.M. (2012) The 2012 nucleic acids research
database issue and the online molecular biology database collection. Nucleic Acids
Res. 40, D1–D8
17 Whitebread, S. et al. (2005) In vitro safety pharmacology profiling: an essential tool
for successful drug development. Drug Discov. Today 10, 1421–1433
18 Vidal, D. et al. (2011) Ligand-based approaches to in silico pharmacology. Methods
Mol. Biol. 672, 489–502
19 Hert, J. et al. (2008) Quantifying the relationships among drug classes. J. Chem. Inf.
Model. 48, 755–765
20 Lagunin, A. et al. (2000) PASS: prediction of activity spectra for biologically active
substances. Bioinformatics 16, 747–748
21 Cheng, T. et al. (2011) Identifying compound–target associations by combining
bioactivity profile similarity search and public databases mining. J. Chem. Inf. Model.
51, 2440–2448
22 Jacoby, E. et al. (2009) Knowledge-based virtual screening: application to the
MDM4/p53 protein–protein interaction. Methods Mol. Biol. 575, 173–194
23 Sayle, R. et al. (2012) Improved chemical text mining of patents with infinite
dictionaries and automatic spelling correction. J. Chem. Inf. Model. 52, 51–62
24 Suriyawongkul, I. et al. (2010) The Cinderella of biological data integration:
addressing some of the challenges of entity and relationship mining from patent
sources. In Data Integration in the Life Sciences (Lambrix, P. and Kemp, G., eds), pp.
106–121, Springer Verlag
25 Fishman, M.C. and Porter, J.A. (2005) Pharmaceuticals: a new grammar for drug
discovery. Nature 437, 491–493
26 Baggs, J.E. (2010) The network as the target. WIREs Syst. Biol. Med. 2, 127–133
27 Pence, D. and Williams, A.J. (2010) ChemSpider: an online chemical information
resource. J. Chem. Educ. 87, 1123–1124
28 Williams, A.J. (2008) Public chemical compound databases. Curr. Opin. Drug Discov.
Dev. 11, 393–404
29 Williams, A.J. and Ekins, S. (2011) A quality alert and call for improved curation of
public chemistry databases. Drug Discov. Today 16, 747–750
30 Mestres, J. et al. (2008) Data completeness – the Achilles heel of drug–target
networks. Nat. Biotechnol. 26, 983–984
31 Schuffenhauer, A. et al. (2002) An ontology for pharmaceutical ligands and its
application for in silico screening and library design. J. Chem. Inf. Comput. Sci. 42,
32 The eTOX Website. Available at: (accessed May 2013)
33 Altman, R.B. et al. (2008) Text mining for biology – the way forward: opinions from
leading scientists. Genome Biol. 9 (Suppl. 2), 7
34 Zimmermann, M. et al. (2005) Information extraction in the life sciences:
perspectives for medicinal chemistry, pharmacology and toxicology. Curr. Top. Med.
Chem. 5, 785–796
35 van Haagen, H. and Mons, B. (2011) In silico knowledge and content tracking.
Methods Mol. Biol. 760, 129–140
36 Industry partnerships. Available at: (accessed May
37 SureChem – Patent chemistry made easy and accessible. Available at: http:// (accessed May 2013)
38 Griffin, T.D. et al. (2010) Annotating patents with Medline MeSH codes via citation
mapping. Adv. Exp. Med. Biol. 680, 737–744
39 Robson, B. et al. (2011) Drug discovery using very large numbers of patents: general
strategy with extensive use of match and edit operations. J. Comput. Aided Mol. Des.
25, 427–441
40 Downs, G.M. and Barnard, J.M. (2011) Chemical patent information systems.
WIREs Comput. Mol. Sci. 1, 727–741
41 Williams, A.J. et al. (2012) Towards a gold standard: regarding quality in public
domain chemistry databases and approaches to improving the situation. Drug
Discov. Today 17, 685–701
42 Muresan, S. et al. (2011) Making every SAR point count: the development of
Chemistry Connect for the large-scale integration of structure and bioactivity data.
Drug Discov. Today 16, 1019–1030
43 Food and Drug Administration Substance Registration System Standard Operating
Procedure. Available at:
ucm127743.pdf (accessed May 2013)
44 Allemang, D. and Hendler, J., eds) (2011) Semantic Web for the Working Ontologist:
Effective Modeling in RDFS and OWL, Morgan Kaufmann
45 Oprea, T.I. et al. (2011) Associating drugs, targets and clinical outcomes into an
integrated network affords a new platform for computer-aided drug repurposing.
Mol. Inform. 30, 100–111
46 Loging, W. et al. (2011) Cheminformatic/bioinformatic analysis of large
corporate databases: application to drug repurposing. Drug Discov. Today 8,
47 Cavalla, D. and Singal, C. (2012) Retrospective clinical analysis for drug rescue: for
new indications or stratified patient groups. Drug Discov. Today 17, 104–109
48 Ekins, S. and Williams, A.J. (2010) Precompetitive preclinical ADME/Tox data: set it
free on the web to facilitate computational model building to assist drug
development. Lab Chip 10, 13–22
49 de Matos, P. et al. (2012) A database for chemical proteomics: ChEBI. Methods Mol.
Biol. 803, 273–296
50 Hastings, J. et al. (2011) The chemical information ontology: provenance and
disambiguation for chemical data on the biological semantic web. PLoS ONE 6,
51 He, B. et al. (2011) Mining relational paths in integrated biomedical data. PLoS ONE
6, e27506
Download PDF