Published online 28 November 2016 Nucleic Acids Research, 2017, Vol.

45, Database issue D945–D954

doi: 10.1093/nar/gkw1074

The ChEMBL database in 2017

Anna Gaulton1 , Anne Hersey1 , Micha-l Nowotka1 , A. Patrı́cia Bento1,2 , Jon Chambers1 ,
David Mendez1 , Prudence Mutowo1 , Francis Atkinson1 , Louisa J. Bellis1 ,
Elena Cibrián-Uhalte1 , Mark Davies1 , Nathan Dedman1 , Anneli Karlsson1 , Marı́a
Paula Magariños1,2 , John P. Overington1 , George Papadatos1 , Ines Smit1 and Andrew
R. Leach1,*
1
European Molecular Biology Laboratory, European Bioinformatics Institute, Wellcome Genome Campus, Hinxton,

Downloaded from https://academic.oup.com/nar/article/45/D1/D945/2605707 by guest on 15 December 2023

Cambridgeshire CB10 1SD, UK and 2 Open Targets, Wellcome Genome Campus, Hinxton, Cambridgeshire CB10
1SD, UK

Received September 23, 2016; Revised October 21, 2016; Editorial Decision October 24, 2016; Accepted October 30, 2016

ABSTRACT feedback, the scope and data types within ChEMBL have
gradually expanded, with some major new areas included
ChEMBL is an open large-scale bioactivity database in recent releases: compounds in clinical development, data
(https://www.ebi.ac.uk/chembl), previously de- from patents, direct depositions for neglected diseases and
scribed in the 2012 and 2014 Nucleic Acids Research agrochemical data.
Database Issues. Since then, alongside the contin- Drug discovery remains a costly process with a high
ued extraction of data from the medicinal chemistry failure rate (3–6). To provide a more complete picture
literature, new sources of bioactivity data have across the drug discovery and development process, and to
also been added to the database. These include: help researchers better understand what makes a success-
deposited data sets from neglected disease screen- ful medicine, we have extended the ChEMBL data model
ing; crop protection data; drug metabolism and to include, for the first time, data typically generated in the
disposition data and bioactivity data from patents. A pre-clinical and clinical phases of drug discovery, specifi-
cally drug metabolism and disposition data. Another com-
number of improvements and new features have also
mon approach to understanding pharmaceutical attrition
been incorporated. These include the annotation of is to learn from successful drugs and failed drug candidates
assays and targets using ontologies, the inclusion (7–9). We have therefore extended our set of drug-target an-
of targets and indications for clinical candidates, notations to include those for clinical candidates and have
addition of metabolic pathways for drugs and calcu- also mapped these chemical entities to their therapeutic in-
lation of structural alerts. The ChEMBL data can be dications.
accessed via a web-interface, RDF distribution, data At the end of 2013, EMBL-EBI took over the operation,
downloads and RESTful web-services. development and support of the SureChem patent system
(now called SureChEMBL (10)) from Digital Science Ltd.
Access to this resource has highlighted the potential value
INTRODUCTION to scientists of bioactivity data not yet published in the sci-
Since its inception a major component of ChEMBL’s con- entific literature. However, the current SureChEMBL sys-
tent has been bioactivity data regularly extracted from the tem only extracts compound structures from the patents
medicinal chemistry literature (1,2). Among many other and not associated bioactivity data. As a first step to ad-
applications such data enables researchers to identify tool dress this opportunity we have worked with BindingDB to
compounds for potential therapeutic targets, to probe the incorporate the BindingDB patent data into ChEMBL (11).
available SAR data for a target, investigate phenotypic data Neglected disease research continues to be a field of drug
associated with similar compounds and to identify potential discovery conducted largely (though not exclusively) by not-
off-target effects of specific chemotypes. In order to pro- for-profit organisations that aim to expedite research by en-
vide a more complete perspective, based in part on user couraging sharing of experimental data with the commu-

* To whom correspondence should be addressed. Tel: +44 1223 494333; Fax: +44 1223 494468; Email: arl@ebi.ac.uk
Present addresses:
Louisa J. Bellis, Department of Genetics, University of Cambridge, Downing Street, Cambridge CB2 3EH, UK.
John P. Overington, Mark Davies and Anneli Karlsson, BenevolentAI, 40 Churchway, London NW1 1LW, UK.
Nathan Dedman, Local Measure, 87 Leonard St, London EC2A 4QS, UK.
George Papadatos, GlaxoSmithKline, Medicines Research Centre, Gunnels Wood Road, Stevenage, Herts SG1 2NY, UK.


C The Author(s) 2016. Published by Oxford University Press on behalf of Nucleic Acids Research.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which
permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
D946 Nucleic Acids Research, 2017, Vol. 45, Database issue

nity (12–15). Depositions of this type of data into ChEMBL Beyond the area of neglected diseases, the University
have continued to increase since the original malaria depo- of Vienna and Roche have deposited supplementary data
sitions in 2010. associated with publications already in ChEMBL (17,18).
Whilst the pharmaceutical and drug discovery commu- This is complementary to the similar sets already de-
nity continues to be the major user and consumer of posited by GlaxoSmithKline and we encourage similar de-
ChEMBL data, other life sciences communities also work positions from other authors. AstraZeneca have taken a
with similar types of data. The agrochemical industry is one different approach to direct data deposition. They iden-
such community where specific efforts have been made to tified compounds already in ChEMBL and then pro-
widen the coverage of data relevant to the discovery and vided data on these compounds from a variety of in
development of herbicides, pesticides and fungicides (16). vitro ADME and physicochemical screens including pro-
In the next sections, we describe the new data types now tein binding, microsome and hepatocyte clearance, solu-
integrated into the ChEMBL database, the annotations we bility, pKa and lipophilicity. It is important to note that
have undertaken to enable structured organisation and ac- for all such deposited data sets, ChEMBL provides a

Downloaded from https://academic.oup.com/nar/article/45/D1/D945/2605707 by guest on 15 December 2023

cess to the data and how the new data and annotations can DOI so that the data can be cited in subsequent publica-
be viewed via the web interface. tions. For example, the DOI for the AstraZeneca deposited
data is 10.6019/CHEMBL3301361 and this resolves to the
DATA CONTENT ChEMBL Document Report Card for the complete data
set.
Current data content Deposited data sets can be identified through the
ChEMBL’s content continues to grow; release 22 of the ChEMBL interface by selecting the relevant entry in the
database contains information extracted from more than ‘Activity Source Filter’ (located to the right of the keyword
65 000 publications, together with 50 deposited data sets, search bar) and then performing a keyword search with a
and data drawn from other databases (Table 1). In total, wildcard (*) against ‘Documents’ or ‘Assays’. This will re-
there are >1.6 million distinct compound structures repre- turn all documents or assays associated with that source or
sented in the database, with 14 million activity values from depositor, from which point further information regarding
>1.2 million assays. These assays are mapped to ∼11 000 compounds, targets and activity measurements can be nav-
targets, including 9052 proteins (of which 4255 are human). igated.

Deposited data sets Crop protection data

ChEMBL continues to receive data sets from both not- In order to broaden the utility of ChEMBL in crop
for-profit and commercial organisations that wish to share protection research, a data set of >40 000 compounds
data with the scientific community. These deposited data and 245 000 activity data points has been extracted
sets contain many novel chemical structures and associated from crop protection-related publications and added to
bioactivity data. A good example is a library of small molec- ChEMBL (16). This data set significantly increased the
ular weight (∼320 Da average) natural product-like com- content of pesticide, herbicide and insecticide assays
pounds created at the University of Dundee with funding in the database. The ChEMBL taxonomy tree browser
from the Gates Foundation. This library is being screened in has been extended to allow easier retrieval of this
a number of neglected disease assays. To date, information data (https://www.ebi.ac.uk/chembl/target/browser). In ad-
about the compound structures and their activity in cyto- dition, known pesticides already in ChEMBL were as-
toxicity assays is available in ChEMBL; further assay data signed a mechanism of action classification, following the
will be deposited as the screening is completed. Another ex- Fungicide Resistance Action Committee (FRAC: http:
ample is the Malaria Box Compound Set (http://www.mmv. //www.frac.info/publications), Herbicide Resistance Ac-
org/research-development/open-access-malaria-box), a set tion Committee (HRAC: http://hrac.tsstaging.com/tools/
of 400 compounds with antimalarial activity that was made classification-lookup) or Insecticide Resistance Action
available by the Medicines for Malaria Venture (MMV) Committee (IRAC: http://www.irac-online.org/documents/
for research groups to request and screen (15). The latest moa-brochure/?ext=pdf) systems.
ChEMBL releases include results from an additional 19
laboratories that have screened and deposited their results
Patent data exchange
on these compounds. The Malaria Box was recently super-
seded by the Pathogen Box (http://www.pathogenbox.org), BindingDB is a database of affinity measurements for small
which is a new set of compounds with activity against a molecules binding to protein targets (11). BindingDB cu-
broader range of pathogens, also available from MMV for rates data from journals complementary to ChEMBL and
screening. These compounds have also been deposited in also incorporates relevant ChEMBL data. More recently,
the ChEMBL database and experimental data will also be the BindingDB team have abstracted binding affinity data
added as it becomes available. Four new sets of screen- from granted US patents. We have worked with BindingDB
ing results from the Drugs for Neglected Diseases Initia- to establish a data exchange mechanism for patent data
tive (DNDi, http://www.dndi.org) have also been incorpo- and now include in ChEMBL the patent data extracted
rated into recent ChEMBL releases, providing data on com- by the BindingDB team (ChEMBL source = 37). Identi-
pounds tested as potential drugs for neglected diseases such fiers for these patent documents are now included in the
as Leishmaniasis and Chagas disease. DOCS table as PATENT ID. Since target information is al-
 Nucleic Acids Research, 2017, Vol. 45, Database issue D947

Table 1. Data sources included in the ChEMBL release 22

No. No. No.
Short name Source compounds assays activities
LITERATURE Scientific Literature 967 242 963 186 5 635 084
PUBCHEM BIOASSAY PubChem BioAssays 489 575 2937 7 559 601
GATES LIBRARY Gates Library compound collection 68 490 2 69 444
BINDINGDB BindingDB Database 68 149 1317 99 061
GSK TCMDC GSK Malaria Screening 13 467 6 81 198
ST JUDE LEISH St Jude Leishmania Screening 13 422 6 42 105
USP/USAN USP Dictionary of USAN and International Drug Names 11 356 0 0
DNDI Drugs for Neglected Diseases Initiative (DNDi) 7053 233 14 452
ASTRAZENECA AstraZeneca Deposited Data 5799 15 11 687
NOVARTIS Novartis Malaria Screening 5614 6 27 888
ORANGE BOOK Orange Book 2016 0 0

Downloaded from https://academic.oup.com/nar/article/45/D1/D945/2605707 by guest on 15 December 2023

SUPPLEMENTARY Deposited Supplementary Bioactivity Data 1786 13 4817
CANDIDATES Clinical Candidates 1633 0 0
ST JUDE St Jude Malaria Screening 1524 16 5456
TP TRANSPORTER TP-search Transporter Database 1434 3592 6765
DRUGMATRIX DrugMatrix 930 113 678 350 929
METABOLISM Curated Drug Metabolism Pathways 828 0 0
GSK TB GSK Tuberculosis Screening 826 15 1814
WHO TDR WHO-TDR Malaria Screening 740 16 5853
GSK TCAKS GSK Kinetoplastid Screening 592 13 7235
MMV MBOX MMV Malaria Box 400 138 45 158
MMV PBOX MMV Pathogen Box 400 0 0
ATLAS Gene Expression Atlas Compounds 398 0 0
DRUGS Manually Added Drugs 378 0 0
GSK PKIS GSK Published Kinase Inhibitor Set 366 456 169 451
OSM Open Source Malaria Screening 211 22 344
WITHDRAWN Withdrawn Drugs 192 0 0
TG GATES Open TG-GATEs 160 158 199 158 199
SANGER Sanger Institute Genomics of Drug Sensitivity in Cancer 137 714 73 169
FDA APPROVAL FDA Approval Packages 43 1386 1387
HARVARD Harvard Malaria Screening 37 4 111

ready carefully manually curated in BindingDB, this infor- the ChEMBL data; it has also been adopted by a number of
mation is retained in ChEMBL and simply mapped to the other bioassay data providers and members of the drug dis-
equivalent ChEMBL target. The data is taken from 1,015 covery community, allowing for good data interoperability.
granted US patents published between 2013 and 2015 and ChEMBL standard activity types were manually mapped
currently comprises 99 061 bioactivities on 68 149 distinct to corresponding BAO result terms (stored in the ACTIVI-
compounds binding to around 600 distinct targets. Of par- TIES table as BAO ENDPOINT). The resulting mappings
ticular interest to drug discovery scientists is the fact that cover 91% of the activity data points in ChEMBL. The re-
data often appears in patents earlier than in the traditional mainder are mainly diverse phenotypic endpoints that are
medicinal chemistry literature. This patent set contains data not covered by BAO (e.g. ‘Tissue Severity Score’, ‘Anticon-
on 50 targets for which there was previously no data in vulsant activity’, ‘Paw swelling’, ‘Relative uterus weight’) or
ChEMBL and which may therefore represent novel targets imprecise terms (e.g. ‘Ratio’, ‘Selectivity’, ‘Response’) that
of therapeutic interest. require further resolution. Similarly, ChEMBL standard
activity units were mapped to Units Ontology (UO) terms
NEW FUNCTIONALITY (22) (which are also a component of BAO) and stored in the
ACTIVITIES table as UO UNITS. Where available, units
Richer assay and target annotation were also mapped to terms from the Quantities, Units, Di-
Typical entry points to ChEMBL have predominantly been mensions and Types ontology (http://www.qudt.org). These
compound-based or target-based searches. However, more are stored in the ACTIVITIES table as QUDT UNITS.
than half of the activity data points in ChEMBL come from The current mapping to Units Ontology covers 87% of
functional or phenotypic assays that cannot be assigned a ChEMBL activity data points (the remainder largely be-
molecular target. Since phenotypic screening is once more ing complex units e.g. ‘ng.h.ml-1 ’, ‘ml.min-1 .g-1 ’ that are not
becoming commonplace in drug discovery (19), making this covered by UO).
wealth of data more accessible is a priority. To this end, we ChEMBL assays have also been annotated with BAO
have applied a number of ontologies to the ChEMBL assay assay format terms, allowing users to distinguish bio-
and activity data, allowing them to be searched and filtered chemical, cell-based, tissue-based or organism-based
by cell-line, tissue or assay format, for example. assays. An automated, rule-based approach was used to
The BioAssay Ontology (BAO) (20,21) was chosen as a classify the assay format for historical assays, based on
means of annotating ChEMBL assays for a number of rea- information in assay descriptions and target assignments.
sons: this ontology has been developed specifically for small In order to minimise false assignments, any assays where
molecule screening data and so provides good coverage of the format could not be determined unambiguously
D948 Nucleic Acids Research, 2017, Vol. 45, Database issue

were left unclassified, pending further curation (11% of

assays). The assay format is stored in the ASSAYS table
as BAO FORMAT. Complementary to this, we have also
introduced annotation of the cells and tissues used in assays
(stored as CELL ID, ASSAY CELL TYPE, TISSUE ID
and ASSAY TISSUE in the ASSAYS table). A ChEMBL
CELL DICTIONARY and TISSUE DICTIONARY have
been introduced and cell and tissue Report Card pages and
search functionality created. It should be noted that the aim
here is not to create new cell-line/tissue ontologies – entries
in the ChEMBL dictionaries are imported from established
vocabularies and ontologies wherever possible (e.g. Cell
Line Ontology (23), Experimental Factor Ontology (EFO)

Downloaded from https://academic.oup.com/nar/article/45/D1/D945/2605707 by guest on 15 December 2023

(24), Cellosaurus (http://web.expasy.org/cellosaurus/)
and LINCS cell dictionary (25) for cells; and Uberon
(26), EFO, Brenda Tissue Ontology (27) and CALOHA
for tissues (ftp://ftp.nextprot.org/pub/current release/
controlled vocabularies/caloha.obo)) and mappings to
these ontologies are provided.
Methods for browsing and retrieving protein target data
in ChEMBL have also been enhanced. Improvements have
been made to the existing protein family classification for
ion channels and transporters to more closely align the Figure 1. Examples of more complex queries that could be performed (e.g.
ChEMBL system with other resources (the IUPHAR/BPS using web services) by combining BioAssay Ontology, protein family and
Guide To PHARMACOLOGY (28) and TCDB (29)) and GO classifications.
new classes have been introduced for epigenetic regulators
(following the ChromoHub database (30)). In order to al-
low browsing of ChEMBL targets by Gene Ontology terms, and clinical candidates for which indication information is
a new GO Slim has been created (31). This is a subset of available, and displays the MeSH and EFO terms for the in-
GO terms that are enriched in ChEMBL targets and allows dications and the highest development phase at which each
the implementation of a simplified Gene Ontology Tree for indication has been investigated. The view can be searched
browsing (on the ‘Browse Targets’ tab). and filtered by typing keywords in the ‘Search’ box (e.g. ‘di-
These new features, particularly when combined, enable abetes’ would filter the display to only those drugs that have
more sophisticated queries of the ChEMBL data to be per- an indication containing the word diabetes). More specific
formed, as illustrated in Figure 1. searches can be performed by entering a MeSH or EFO ID
in the search box (e.g. ‘EFO:0001360’ would return drugs
indicated for ‘type II diabetes mellitus’). The whole or fil-
Drug indications and clinical candidates
tered data set can also be exported using the ‘Downloads’
In previous releases of ChEMBL we annotated proper- option at the top right-hand side of the view. A summary of
ties and efficacy targets for FDA approved drugs. We are the indication data is also displayed on Compound Report
now also incorporating drug indication information into Card pages for drugs.
ChEMBL to facilitate use-cases such as target validation Furthermore, we are now extending the annotation of ef-
(the existence of a drug that acts through a particular tar- ficacy targets and indications to cover drug candidates in
get and is prescribed for a given indication may be taken clinical development. Mechanism of action and target in-
as evidence validating the relevance of that target for the formation for 1,023 clinical candidates in phase I-III devel-
indication). Drug indication data is collected from a vari- opment (mainly targeting GPCR, kinase, nuclear hormone
ety of sources including DailyMed package inserts (https:// receptor and ion channel targets, which are the focus of the
dailymed.nlm.nih.gov), Anatomical Therapeutic Chemical NIH-funded Illuminating the Druggable Genome project:
(ATC) classification (http://www.whocc.no/atc ddd index/) https://commonfund.nih.gov/idg/) has already been curated
(32) and ClinicalTrials.gov (https://clinicaltrials.gov) (33). and included in the database (ChEMBL source = 8), and
Since these resources mainly provide free text information, a targets for further candidates will be included in future re-
combination of text-mining (using BeCAS (34)), automated leases. Indications for candidates have also been included
mapping and manual curation/validation is used to iden- using information in ClinicalTrials.gov. For each candidate-
tify the indication terms and assign the corresponding dis- indication association, the highest development phase at
ease terms in the Medical Subject Headings (MeSH) vo- which that indication has been investigated is also recorded.
cabulary (https://www.nlm.nih.gov/mesh/) and EFO (24). Currently 1,451 clinical candidates have associated indica-
Two new tables have been added to the database schema tion information. Clinical candidates are now included in
to capture this information: DRUG INDICATION and the ‘Browse Drugs’ tab on the ChEMBL interface and their
INDICATION REFS. A new tab has been added to the mechanism of action and indication information (where an-
ChEMBL interface to display the data––‘Browse Drug In- notated) can be viewed in the ‘Browse Drug Targets’ and
dications’. This view contains a list of all approved drugs ‘Browse Drug Indications’ tabs, respectively.
 Nucleic Acids Research, 2017, Vol. 45, Database issue D949

Finally, we have also added information regard- in a data model that maintains the relationships between the
ing previously approved drugs that have been with- various molecular entities (e.g. metabolite A may be formed
drawn for toxicity or efficacy reasons. Information directly from drug D, whereas metabolite B may result
regarding withdrawn drugs was collated from several from the degradation of metabolite A). Where metabolis-
sources: the FDA (http://www.fda.gov) and EMA (http: ing enzymes, species and tissues are available in the origi-
//www.ema.europa.eu/ema/), the WITHDRAWN database nal publication this information is recorded in ChEMBL.
(35), the US Electronic Code of Federal Regulations In instances where the metabolite structure is known it is
(http://www.ecfr.gov/cgi-bin/retrieveECFR?gp=2&SID= recorded as a chemical structure. If the exact structure is
915cc9ab8176f1d1a2a355acf064ffe3&h=L&mc=true&n= unknown the reaction is still recorded but with an unde-
sp21.4.216.b&r=SUBPART&ty=HTML#se21.4.216 124), fined structure. The metabolism data is recorded in two
Federal Register (https://www.gpo.gov/fdsys/pkg/FR- new tables METABOLISM and METABOLISM REFS.
2014-07-02/pdf/2014-15371.pdf) and several review ar- The metabolite pathway is shown on the ChEMBL in-
ticles (36–38). Where available, the year of withdrawal, terface as an interactive image with links to the data on

Downloaded from https://academic.oup.com/nar/article/45/D1/D945/2605707 by guest on 15 December 2023

the applicable countries/areas and the reasons for the the metabolites, the metabolising enzymes and the docu-
withdrawal are captured. This information has been added ment sources for the information. An example for Sim-
to the MOLECULE DICTIONARY database table and vastatin (https://www.ebi.ac.uk/chembl/compound/inspect/
is displayed on the ‘Browse Drugs’ tab and also on Com- CHEMBL1064) is shown in Figure 3.
pound Report Card pages (see Figure 2). The ‘Molecule
Features’ icons have also been updated to include a new Compound structural alerts
availability type icon. This icon now has four options: with-
drawn, discontinued, prescription-only, over-the-counter. In order to aid users in the selection of compounds,
In cases where the drug has been withdrawn in one country for example to create screening sets, we have compiled
but is still available in others, the icon will retain the sets of publicly-available structural alerts as SMARTS
prescription-only or over-the-counter status, rather than (Pfizer LINT filters (42), Glaxo Wellcome Hard Fil-
withdrawn. ters (43), Bristol-Myers Squibb HTS Deck Filters
(44), NIH MLSMR Excluded Functionality Filters
(https://mlsmr.evotec.com/MLSMR HomePage/pdf/
Drug metabolism, toxicity and pharmacokinetic data MLSMR Excluded Functionality Filters 200605121510.
In vivo data can be extremely valuable but is also invari- pdf), University of Dundee NTD Screening Library
ably complex, with experimental parameters typically being Filters (45) and Pan Assay Interference Compounds
measured at varying doses, time points, routes of admin- (PAINS) Filters (46), filters derived by Inpharmat-
istration, in the presence of interacting compounds etc. In ica Ltd, set of alerts currently used in SureChEMBL
order to organise this information in a more structured way https://www.surechembl.org/knowledgebase/169485).
and to enable easier access of this information by ChEMBL All filters have been run against all ChEMBL
users an ASSAY PARAMETERS table has been included compounds for which structures are available and
in the ChEMBL schema to record this information. This the results included in the database in the COM-
has been done retrospectively for all the ChEMBL assays POUND STRUCTURAL ALERTS, STRUC-
and resulted in addition of 3.6 million parameter mappings TURAL ALERTS and STRUCTURAL ALERT SETS
for approximately one third of all the ChEMBL assays. tables. These filters can be used to identify compounds
Additionally, new information has been added to that may be problematic in a drug-discovery setting and
ChEMBL on drug metabolism, pharmacokinetics and tox- so guide the selection of compounds and/or help in the
icology particularly for pre-clinical and clinical drug can- interpretation of results. The filters typically represent sub-
didates. The data has been manually extracted from three structures corresponding to chemically reactive functional
new data sources and will be extended in future ChEMBL groups, those that are associated with toxicity, interfere
releases. Firstly ∼260 articles covering the years 2011–2013 with certain assay formats or bind promiscuously to targets.
were extracted from the Journal of Drug Metabolism and Alerts are implemented as reported in the original source
Disposition. This resulted in data on ∼2200 compounds but should be interpreted with care, depending on the
tested in nearly 15 000 assays (included in ChEMBL source use-case, and not treated as a blanket filter (e.g. ∼50% of
= 1). The second data set included are the FDA drug ap- approved drugs have one or more alerts from these sets).
proval packages (39). These are summaries of the informa- These alerts can be viewed on the Compound Report Card
tion submitted to the FDA when the drug was approved for via the ChEMBL interface. The alerts are organised by
use and are available for FDA drugs approved after 1997. the alert set that a specific functionality is found in and
To our knowledge this valuable data is not captured in a the functional group is highlighted on a depiction of the
structured format elsewhere (ChEMBL source = 28). molecule.
Thirdly, given the need to understand the nature of drug
metabolites and whether they themselves are therapeuti- DATA ACCESS
cally active or constitute reactive species that bind DNA or
The ChEMBL interface
proteins resulting in toxicity (40,41), a project has been ini-
tiated to identify the metabolic pathways of approved drugs. The ChEMBL database is accessible via a basic user inter-
This was achieved using a variety of data sources (ChEMBL face at: https://www.ebi.ac.uk/chembl. This interface pro-
sources 1, 28 and 31). The metabolic schemes are described vides core search functionality for compounds (including
D950 Nucleic Acids Research, 2017, Vol. 45, Database issue

Downloaded from https://academic.oup.com/nar/article/45/D1/D945/2605707 by guest on 15 December 2023

Figure 2. Compound Report Card for Troglitazone showing mechanism of action, indication and withdrawal information (https://www.ebi.ac.uk/chembl/
compound/inspect/CHEMBL408).
 Nucleic Acids Research, 2017, Vol. 45, Database issue D951

Downloaded from https://academic.oup.com/nar/article/45/D1/D945/2605707 by guest on 15 December 2023

Figure 3. Metabolism scheme for Simvastatin (https://www.ebi.ac.uk/chembl/compound/inspect/CHEMBL1064).

D952 Nucleic Acids Research, 2017, Vol. 45, Database issue

substructure and similarity searching) and targets as well amount of data that can be retrieved for a particu-
as keyword searching across assay, cell line and tissue infor- lar target/compound, or to provide relevant pharmacol-
mation. Users can also retrieve and filter bioactivity infor- ogy and drug-target data in the context of other data
mation and browse drug and clinical candidate information types (e.g. pathway, expression or disease information).
(including targets and indications). More details of the user ChEMBL data is incorporated into a wide range of other
interface and its functionality can be found in previous pub- resources including PubChem BioAssay (52), BindingDB
lications (1,2). (11), CanSAR (53), Open PHACTS (54), Open Targets (55)
and the Target Central Resource Database/PHAROS (http:
Downloads and web-services //juniper.health.unm.edu/tcrd/, 56), so can also be accessed
via these routes. However, since these other resources are
While the ChEMBL interface provides the functionality
different in scope, they do not all incorporate ChEMBL in
necessary for many simple use-cases, some users may pre-
full (e.g. BindingDB focuses only on binding measurements,
fer to download the database and query it locally (e.g. for
while Open Targets incorporates data on drug–target and

Downloaded from https://academic.oup.com/nar/article/45/D1/D945/2605707 by guest on 15 December 2023

use in data mining applications or to integrate with inter-
drug–indication linkage).
nal data). Each release of ChEMBL is available from our
ftp site in a variety of formats including: Oracle, MySQL,
PostgreSQL, SQLite, RDF, an SD file of compound struc- SUMMARY
tures and a FASTA file of the target sequences, under a
Creative Commons Attribution-ShareAlike 3.0 Unported ChEMBL continues to evolve and grow in order to ad-
license (http://creativecommons.org/licenses/by-sa/3.0). dress an ever-increasing range of use-cases, user communi-
The myChEMBL software container is also available ties and applications. The ‘core’ of the resource continues to
(47,48). It is distributed as a virtual machine disk im- be molecule–target interaction data curated from the pub-
age in many formats such as VMDK, QCOW2 and IMG lished literature, but as outlined above and in previous re-
that can be downloaded using FTP or Vagrant. Alterna- views (1,2,51,57,58) this has been expanded both in terms of
tively, myChEMBL can be used as a Docker container. data content, annotation and infrastructure. Many of these
The container is based on Ubuntu OS but there exists developments have been in response to feedback and dis-
a CentOS version as well. The container consists of the cussion from ChEMBL’s growing user community; we wel-
ChEMBL PostgreSQL database, web services, open source come such engagement on the latest version of the resource.
cheminformatics tools (e.g. RDKit: http://www.rdkit.org,
OSRA (49) and Open Babel (50)) and Jupyter Note- ACKNOWLEDGEMENTS
book tutorials (https://github.com/chembl/mychembl/tree/
master/ipython notebooks), providing a convenient envi- Molecular Connections, Mike Gilson, Tiqing Liu, Kather-
ronment for users to interrogate the data. ine Doyle.
For users requiring programmatic access to ChEMBL, a
comprehensive set of RESTful web services are provided
(51), allowing retrieval of ChEMBL data in XML, JSON FUNDING
and YAML formats (see https://www.ebi.ac.uk/chembl/ws Strategic Awards from the Wellcome Trust
for more details and example queries). A Solr-based [WT086151/Z/08/Z, WT104104/Z/14/Z]; Member
search functionality is also now available, permitting States of the European Molecular Biology Laboratory
retrieval of compound, target and assay information (EMBL); Innovative Medicines Initiative Joint Undertak-
via text queries. For example, the following query ing [115191]; National Institutes of Health (NIH) Common
would retrieve all assays containing the term ‘angio- Fund [U54CA189205]; European Union Seventh Frame-
genesis’: https://www.ebi.ac.uk/chembl/api/data/assay/ work Programme (FP7/2007–2013) [602156]; Syngenta;
search?q=angiogenesis. An example Python web service Open Targets. Funding for open access charge: Wellcome
client library is available: https://github.com/chembl/ Trust.
chembl webresource client, with usage examples covered Conflict of interest statement. None declared.
in: https://github.com/chembl/mychembl/blob/master/
ipython notebooks/09 myChEMBL web services.ipynb.
Web services and myChEMBL are both open source REFERENCES
projects and are available from the ChEMBL GitHub 1. Bento,A.P., Gaulton,A., Hersey,A., Bellis,L.J., Chambers,J.,
repository (https://github.com/chembl/), along with other Davies,M., Kruger,F.A., Light,Y., Mak,L., McGlinchey,S. et al.
useful resources to help integration of ChEMBL into soft- (2014) The ChEMBL bioactivity database: an update. Nucleic Acids
Res., 42, D1083–D1090.
ware projects (https://arxiv.org/pdf/1607.00378.pdf). They 2. Gaulton,A., Bellis,L.J., Bento,A.P., Chambers,J., Davies,M.,
are licensed under an Apache 2 license. Both projects con- Hersey,A., Light,Y., McGlinchey,S., Michalovich,D., Al-Lazikani,B.
verge to build a new version of the main ChEMBL web in- et al. (2012) ChEMBL: a large-scale bioactivity database for drug
terface which will be built on top of web services and de- discovery. Nucleic Acids Res., 40, D1100–D1107.
3. Arrowsmith,J. and Miller,P. (2013) Trial watch: phase II and phase III
ployed and distributed using the myChEMBL container. attrition rates 2011-2012. Nat. Rev. Drug Discov., 12, 569.
4. Bunnage,M.E. (2011) Getting pharmaceutical R&D back on target.
Inclusion of ChEMBL data in other resources Nat. Chem. Biol., 7, 335–339.
5. Hay,M., Thomas,D.W., Craighead,J.L., Economides,C. and
It is often advantageous to integrate ChEMBL data Rosenthal,J. (2014) Clinical development success rates for
with data from other resources––either to maximise the investigational drugs. Nat. Biotechnol., 32, 40–51.
 Nucleic Acids Research, 2017, Vol. 45, Database issue D953

6. Kola,I. and Landis,J. (2004) Can the pharmaceutical industry reduce data from the library of integrated network-based cellular signatures
attrition rates? Nat. Rev. Drug Discov., 3, 711–715. (LINCS). J. Biomol. Screen., 19, 803–816.
7. Cook,D., Brown,D., Alexander,R., March,R., Morgan,P., 26. Mungall,C.J., Torniai,C., Gkoutos,G.V., Lewis,S.E. and
Satterthwaite,G. and Pangalos,M.N. (2014) Lessons learned from the Haendel,M.A. (2012) Uberon, an integrative multi-species anatomy
fate of AstraZeneca’s drug pipeline: a five-dimensional framework. ontology. Genome Biol., 13, R5.
Nat. Rev. Drug Discov., 13, 419–431. 27. Gremse,M., Chang,A., Schomburg,I., Grote,A., Scheer,M.,
8. Waring,M.J., Arrowsmith,J., Leach,A.R., Leeson,P.D., Mandrell,S., Ebeling,C. and Schomburg,D. (2011) The BRENDA Tissue Ontology
Owen,R.M., Pairaudeau,G., Pennie,W.D., Pickett,S.D., Wang,J. et al. (BTO): the first all-integrating ontology of all organisms for enzyme
(2015) An analysis of the attrition of drug candidates from four major sources. Nucleic Acids Res., 39, D507–D513.
pharmaceutical companies. Nat. Rev. Drug Discov., 14, 475–486. 28. Southan,C., Sharman,J.L., Benson,H.E., Faccenda,E., Pawson,A.J.,
9. Morgan,P., Van Der Graaf,P.H., Arrowsmith,J., Feltner,D.E., Alexander,S.P., Buneman,O.P., Davenport,A.P., McGrath,J.C.,
Drummond,K.S., Wegner,C.D. and Street,S.D. (2012) Can the flow Peters,J.A. et al. (2016) The IUPHAR/BPS Guide to
of medicines be improved? Fundamental pharmacokinetic and PHARMACOLOGY in 2016: towards curated quantitative
pharmacological principles toward improving Phase II survival. Drug interactions between 1300 protein targets and 6000 ligands. Nucleic
Discov. Today, 17, 419–424. Acids Res., 44, D1054–D1068.

Downloaded from https://academic.oup.com/nar/article/45/D1/D945/2605707 by guest on 15 December 2023

10. Papadatos,G., Davies,M., Dedman,N., Chambers,J., Gaulton,A., 29. Saier,M.H. Jr, Reddy,V.S., Tsu,B.V., Ahmed,M.S., Li,C. and
Siddle,J., Koks,R., Irvine,S.A., Pettersson,J., Goncharoff,N. et al. Moreno-Hagelsieb,G. (2016) The Transporter Classification Database
(2016) SureChEMBL: a large-scale, chemically annotated patent (TCDB): recent advances. Nucleic Acids Res., 44, D372–D379.
document database. Nucleic Acids Res., 44, D1220–D1228. 30. Shah,M.A., Denton,E.L., Liu,L. and Schapira,M. (2014)
11. Gilson,M.K., Liu,T., Baitaluk,M., Nicola,G., Hwang,L. and ChromoHub V2: cancer genomics. Bioinformatics, 30, 590–592.
Chong,J. (2016) BindingDB in 2015: A public database for medicinal 31. Mutowo,P., Bento,A.P., Dedman,N., Gaulton,A., Hersey,A.,
chemistry, computational chemistry and systems pharmacology. Lomax,J. and Overington,J.P. (2016) A drug target slim: using Gene
Nucleic Acids Res., 44, D1045–D1053. Ontology and Gene Ontology annotations to navigate protein-ligand
12. Gamo,F.J., Sanz,L.M., Vidal,J., de Cozar,C., Alvarez,E., target space in ChEMBL. J. Biomed. Semantics, 7, 59.
Lavandera,J.L., Vanderwall,D.E., Green,D.V., Kumar,V., Hasan,S. 32. WHO Collaborating Centre for Drug Statistics Methodology. (2015)
et al. (2010) Thousands of chemical starting points for antimalarial Guidelines for ATC classification and DDD assignment 2015. WHO
lead identification. Nature, 465, 305–310. Collaborating Centre for Drug Statistics Methodology, Oslo.
13. Ioset,J.R. and Chang,S. (2011) Drugs for Neglected Diseases 33. Huston,M.W., Williams,R.J., Bergeris,A., Fun,J. and Tse,T. (2012)
initiative model of drug development for neglected diseases: current New style and new content for ClinicalTrials.gov. NLM Tech. Bull.,
status and future challenges. Future Med. Chem., 3, 1361–1371. 387, e5.
14. Orti,L., Carbajo,R.J., Pieper,U., Eswar,N., Maurer,S.M., Rai,A.K., 34. Nunes,T., Campos,D., Matos,S. and Oliveira,J.L. (2013) BeCAS:
Taylor,G., Todd,M.H., Pineda-Lucena,A., Sali,A. et al. (2009) A biomedical concept recognition services and visualization.
kernel for open source drug discovery in tropical diseases. PLoS Negl. Bioinformatics, 29, 1915–1916.
Trop. Dis., 3, e418. 35. Siramshetty,V.B., Nickel,J., Omieczynski,C., Gohlke,B.O.,
15. Spangenberg,T., Burrows,J.N., Kowalczyk,P., McDonald,S., Drwal,M.N. and Preissner,R. (2016) WITHDRAWN–a resource for
Wells,T.N. and Willis,P. (2013) The open access malaria box: a drug withdrawn and discontinued drugs. Nucleic Acids Res., 44,
discovery catalyst for neglected diseases. PLoS One, 8, e62906. D1080–D1086.
16. Gaulton,A., Kale,N., van Westen,G.J., Bellis,L.J., Bento,A.P., 36. Bakke,O.M., Manocchia,M., de Abajo,F., Kaitin,K.I. and
Davies,M., Hersey,A., Papadatos,G., Forster,M., Wege,P. et al. (2015) Lasagna,L. (1995) Drug safety discontinuations in the United
A large-scale crop protection bioassay data set. Sci. Data, 2, 150032. Kingdom, the United States, and Spain from 1974 through 1993: a
17. Ceccarelli,S.M., Chomienne,O., Gubler,M. and Arduini,A. (2011) regulatory perspective. Clin. Pharmacol. Ther., 58, 108–117.
Carnitine palmitoyltransferase (CPT) modulators: a medicinal 37. Qureshi,Z.P., Seoane-Vazquez,E., Rodriguez-Monguio,R.,
chemistry perspective on 35 years of research. J. Med. Chem., 54, Stevenson,K.B. and Szeinbach,S.L. (2011) Market withdrawal of new
3109–3152. molecular entities approved in the United States from 1980 to 2009.
18. Kaiser,D., Terfloth,L., Kopp,S., Schulz,J., de Laet,R., Chiba,P., Pharmacoepidemiol. Drug Saf., 20, 772–777.
Ecker,G.F. and Gasteiger,J. (2007) Self-organizing maps for 38. Fung,M., Thornton,A., Mybeck,K., Wu,J.H., Hornbuckle,K. and
identification of new inhibitors of P-glycoprotein. J. Med. Chem., 50, Muniz,E. (2001) Evaluation of the characteristics of safety
1698–1702. withdrawal of prescription drugs from worldwide pharmaceutical
19. Lee,J.A., Uhlik,M.T., Moxham,C.M., Tomandl,D. and Sall,D.J. markets-1960 to 1999. Drug Inf. J., 35, 293–317.
(2012) Modern phenotypic drug discovery is a viable, neoclassic 39. Turner,E.H. (2013) How to access and process FDA drug approval
pharma strategy. J. Med. Chem., 55, 4527–4538. packages for use in research. BMJ, 347, f5992.
20. Abeyruwan,S., Vempati,U.D., Kucuk-McGinty,H., Visser,U., 40. Foti,R.S. and Dalvie,D.K. (2016) Cytochrome P450 and
Koleti,A., Mir,A., Sakurai,K., Chung,C., Bittker,J.A., Clemons,P.A. Non-Cytochrome P450 oxidative metabolism: Contributions to the
et al. (2014) Evolving BioAssay Ontology (BAO): modularization, pharmacokinetics, safety, and efficacy of xenobiotics. Drug Metab.
integration and applications. J. Biomed. Semantics, 5, S5. Dispos., 44, 1229–1245.
21. Visser,U., Abeyruwan,S., Vempati,U., Smith,R.P., Lemmon,V. and 41. Kalgutkar,A.S., Gardner,I., Obach,R.S., Shaffer,C.L., Callegari,E.,
Schurer,S.C. (2011) BioAssay Ontology (BAO): a semantic Henne,K.R., Mutlib,A.E., Dalvie,D.K., Lee,J.S., Nakai,Y. et al.
description of bioassays and high-throughput screening results. BMC (2005) A comprehensive listing of bioactivation pathways of organic
Bioinformatics, 12, 257. functional groups. Curr. Drug Metab., 6, 161–225.
22. Gkoutos,G.V., Schofield,P.N. and Hoehndorf,R. (2012) The Units 42. Blake,J.F. (2005) Identification and evaluation of molecular
Ontology: a tool for integrating units of measurement in science. properties related to preclinical optimization and clinical fate. Med.
Database (Oxford), 2012, bas033. Chem., 1, 649–655.
23. Sarntivijai,S., Lin,Y., Xiang,Z., Meehan,T.F., Diehl,A.D., 43. Hann,M., Hudson,B., Lewell,X., Lifely,R., Miller,L. and
Vempati,U.D., Schurer,S.C., Pang,C., Malone,J., Parkinson,H. et al. Ramsden,N. (1999) Strategic pooling of compounds for
(2014) CLO: The cell line ontology. J. Biomed. Semantics, 5, 37. high-throughput screening. J. Chem. Inf. Comput. Sci., 39, 897–902.
24. Malone,J., Holloway,E., Adamusiak,T., Kapushesky,M., Zheng,J., 44. Pearce,B.C., Sofia,M.J., Good,A.C., Drexler,D.M. and Stock,D.A.
Kolesnikov,N., Zhukova,A., Brazma,A. and Parkinson,H. (2010) (2006) An empirical process for the design of high-throughput
Modeling sample variables with an experimental factor ontology. screening deck filters. J. Chem. Inf. Model., 46, 1060–1068.
Bioinformatics, 26, 1112–1118. 45. Brenk,R., Schipani,A., James,D., Krasowski,A., Gilbert,I.H.,
25. Vempati,U.D., Chung,C., Mader,C., Koleti,A., Datar,N., Vidovic,D., Frearson,J. and Wyatt,P.G. (2008) Lessons learnt from assembling
Wrobel,D., Erickson,S., Muhlich,J.L., Berriz,G. et al. (2014) screening libraries for drug discovery for neglected diseases.
Metadata standard and data exchange specifications to describe, ChemMedChem, 3, 435–444.
model, and integrate complex and diverse high-throughput screening 46. Baell,J.B. and Holloway,G.A. (2010) New substructure filters for
removal of pan assay interference compounds (PAINS) from
D954 Nucleic Acids Research, 2017, Vol. 45, Database issue

screening libraries and for their exclusion in bioassays. J. Med. research and drug discovery knowledgebase. Nucleic Acids Res., 44,
Chem., 53, 2719–2740. D938–D943.
47. Ochoa,R., Davies,M., Papadatos,G., Atkinson,F. and Overington,J.P. 54. Williams,A.J., Harland,L., Groth,P., Pettifer,S., Chichester,C.,
(2014) myChEMBL: a virtual machine implementation of open data Willighagen,E.L., Evelo,C.T., Blomberg,N., Ecker,G., Goble,C. et al.
and cheminformatics tools. Bioinformatics, 30, 298–300. (2012) Open PHACTS: semantic interoperability for drug discovery.
48. Davies,M., Nowotka,M., Papadatos,G., Atkinson,F., van Drug Discov. Today, 17, 1188–1198.
Westen,G.J., Dedman,N., Ochoa,R. and Overington,J.P. (2014) 55. Koscielny,G., An,P., Carvalho-Silva,D., Cham,J.A., Munoz-Pomer
MyChEMBL: a virtual platform for distributing cheminformatics Fuentes,A., Fumis,L., Gasparyan,R., Hasan,S., Karamanis,N.,
tools and open data. Challenges, 5, 334–337. Maguire,M. et al. (2016) Open targets: a platform for therapeutic
49. Filippov,I.V. and Nicklaus,M.C. (2009) Optical structure recognition target identification and validation. Nucleic Acids Res.,
software to recover chemical information: OSRA, an open source doi:10.1093/nar/gkw1055.
solution. J. Chem. Inf. Model., 49, 740–743. 56. Nguyen,D.T., Mathias,D., Bologa,C., Brunak,S., Fernandez,N.,
50. O’Boyle,N.M., Banck,M., James,C.A., Morley,C., Vandermeersch,T. Gaulton,A., Hersey,A., Holmes,J., Jensen,L., Karlsson,A. et al.
and Hutchison,G.R. (2011) Open babel: an open chemical toolbox. J. (2016) Pharos: collating protein information to shed light on the
Cheminform., 3, 33. druggable genome. Nucleic Acids Res., doi:10.1093/nar/gkw1072.

Downloaded from https://academic.oup.com/nar/article/45/D1/D945/2605707 by guest on 15 December 2023

51. Davies,M., Nowotka,M., Papadatos,G., Dedman,N., Gaulton,A., 57. Chambers,J., Davies,M., Gaulton,A., Papadatos,G., Hersey,A. and
Atkinson,F., Bellis,L. and Overington,J.P. (2015) ChEMBL web Overington,J.P. (2014) UniChem: extension of InChI-based
services: streamlining access to drug discovery data and utilities. compound mapping to salt, connectivity and stereochemistry layers.
Nucleic Acids Res., 43, W612–W620. J. Cheminform., 6, 43.
52. Wang,Y., Cheng,T., Wang,J., Gindulyte,A., Shoemaker,B.A., 58. Papadatos,G., Gaulton,A., Hersey,A. and Overington,J.P. (2015)
Thiessen,P.A., He,S., Zhang,J. and Bryant,S.H. (2016) PubChem Activity, assay and target data curation and quality in the ChEMBL
BioAssay: 2017 update. Nucleic Acids Res., database. J. Comput. Aided Mol. Des., 29, 885–896.
doi:10.1093/nar/gkw1118.
53. Tym,J.E., Mitsopoulos,C., Coker,E.A., Razaz,P., Schierz,A.C.,
Antolin,A.A. and Al-Lazikani,B. (2016) canSAR: an updated cancer