Drug News Perspect 16(2), March 2003© 2003 Prous Science CCC: 214–0934/200393
by Edgar Jacoby,
Ansgar Schuffenhauer
and Philipp Floersheim
T
o organize drug discovery
around target families is not a
new concept. Indeed, in the past
a number of target families, e.g., the
monoamine GPCRs, were systematic-
ally explored in a way that selective
ligands are today being known for a
large number of receptors of these
families. Discoveries of new binding
sites for known hormones or drugs
marked the early days of molecular
pharmacology, later followed by the
identification of the corresponding
molecular receptors, which were in
many cases subtypes or sub-subtypes
of previously investigated targets.
Lead finding and lead optimization
programs, searching selective func-
tional ligands and drugs for such new
targets were then initiated in many
cases from the known related reference
compounds.
The elucidation of the human
genome challenged this approach by
the fact that now all members
(sequences) of a target family are visi-
ble and accessible to begin with. The
functional characte ization of new tar-
gets, including in many cases the iden-
tific tio of th endoge ous interaction
partn rs, mi ht nly follo afterwards
in this revers phar acology set-up.
1
With this aim, potent and specific func-
tional molecules arising from chem-
istry are potentially the most efficient
resource for validating the novel tar-
gets.
2
The chances that such a strategy
will further impact the drug discovery
process are very high. Ind ed, because
of the simil rities existing within a
target fa il or an homogenous sub-
group th reof, specially f r aspects of
mol cular recognition, it is only logical
that through the further focus within
target families it will be possible to dis-
cover ligands and drugs of new targets
more quickly and o improve the inno-
vation defi it of the pharmaceutical
industry.
2,3Within this context, chemo-
geno cs, the further systematization
LOOKING AHEAD
The fact that similar ligands bind to similar targets is the underlying principle
of chemogenomics knowledge-based strategies for drug discovery.
Chemogenomics Knowledge-Based
Strategies in Drug Discovery
Summary
In the postgenomic age of drug discovery, targets can no longer be viewed as
singular objects having no relationship to one another. All targets are now visible and
the syst matic exploration of selected target families appears to be a promising way
to speed up and further industrialize target-based drug discovery. Chemogenomics
refers to such systematic exploration of target families and aims to identify all possi-
ble ligands of all target families. Because biology works by applying prior knowledge
to an nknown entity, chemogenomics approaches are expected to be especially
effective within the previously well-explored target families, for which, in addition to
the protein sequence nd structure information, considerable knowledge of pharma-
cologically ct ve tructural classes and structure-activity relationships exists. For the
new target familie , chemical knowledge will have to be generated and beyond bio-
logical target validation, he emphasis is on chemistry to provide the molecules with
which their ovel biology and pharmacology can be studied. Using examples from the
previously most uccessfully explored target families, the GPCR family in particular,
w summarize h rein our current chemogenomics knowledge-based strategies for
drug discovery, which are founded on the high integration of chem and bioinforma-
tics, thereby providing a molecular informatics frame for the exploration of the new
target families.© 2003 Prous Scienc . All rights reserved.

and intensification of the target class
approach, emerged as a drug discovery
paradigm that aims to identify all pos-
sible chemical ligands and drugs of all
target families.
4,5
This report will sum-
marize current chemogenomics strate-
gies for drug discovery. As the entire
process is knowledge driven, the
emphasis will be on the integration of
chem and bioinformatics into a molec-
ular informatics platform for drug dis-
covery in general and, more specifical-
ly, for lead finding. The report is
structured in four parts: 1) a molecular
information system for the pharmaceu-
tical ligands of the main target fami-
lies; 2) bioinformatics identification of
target subfamilies with conserved mol-
ecular recognition; 3) cheminforma-
tics homology-based similarity search-
ing; and 4) knowledge-based ligand
design strategies within homogenous
target subfamilies. Other obvious
advantages of target family approach-
es, such as the homogeneity of assay
development and HTS technologies,
will not be discussed in this report.
A molecular information
system for pharmaceutical
ligands of the main target
families
The newly identified macromolec-
ular targets may belong in part to estab-
lished therapeutically important target
classes like enzymes, G-protein-cou-
pled receptors (GPCRs), nuclear recep-
tors (NRs) and ligand-gated ion chan-
nels (LGICs), which are the most
successful drug target families and
which are early examples of the sys-
tematization approach.
5
Correspond-
ingly, every newly discovered orphan
receptor of these classes can be con-
sidered as a potential drug target.
6
Considering the broad knowledge
existing about the previously investi-
gated members of these families,
including the structural classes of phar-
macologically active compounds and
the sequence information, it is logical
to expect that the pharmacological
investigation of the new targets should
benefit from knowledge-based com-
pound selection and design strategies
that try to extract relevant characteris-
tics from the established knowledge.
To realize this expec ation, given that
the ch m and bioinformatics worlds
have evolved more or less indepen-
dently, it is necessary t establish cross
references between chem and bioinfor-
matics by appr priate annotation
schemes. Annotation fforts in bio-
sciences in the past have focused main-
ly on the annotation of genomic
sequences and comprehensive gene
on ologies like GO,
7
annotating the
biological process, the molecular func-
tion and the cellular component of
gene p oducts. More specifically, sev-
eral no enclature and classification
committees have organized compre-
hensive cla s-specific molecular infor-
mation systems for enzymes,
8
GPCRs,
9,10
N
10
and LGICs.
11
In
comparison, on y very limited effort
has been put into annotation schemes
for ligands. Ligand molecular informa-
tion systems have mainly evolved from
the need to track literature, patent and
clinical status inform tion. Catalogs,
like MDDR,12WDI,13CMC,14dDB15
or PharmaProjects
16
e typical data-
base systems which provide structural
info mati n about ligands together
with molecular target or therapeutic
class information. As he molecular
target information provided within the
ligand database systems contains only
the targ t na e, if that, and does not
provide any further relationship among
th targets, the p ntial of these sys-
ems for lead finding applications
remains limited. Ligands of close
homologous receptors are generally
accepted as putativ starting points in
l d finding programs for receptors for
which no specific ligands are yet
known.
17
Therefore ligand classifica-
tion chemes that reflect phylogenetic
or oth r re ationships of conserved
molecular recognition should be useful
f r lead finding. Correspondingly, we
recently described th adaptation of
annotation schemes f r lig nds of four
major target families of i terest to us
(Fig. 1).
18
The MDDR01.112dat base,
which includes target information for a
large numb r of its ligands, constituted
the underlying ligand dataset. The lig-
and-target classification for the four
target families is b sed on the refer-
enc established by the EC,
8
GPCRDB,
10
NuclearDB,
10
and
LIGCDB.
11
By l king MDDR activi-
ty k ys to the targets of the classifica-
tion schemes, we were able to group
th MDDR ligands within their macro-
molecular arget cl sses. In total, 309
of the 799 ac vity keys used in
MDDR01.1 could be linked to a target,
which allowed us to annotate 53211 of
the total 113821 compounds.
The main purpose of o r ligand
ontology is that liga ds of specified
levels can easily be collated to serve as
comprehe sive reference sets for
cheminform tics-based similarity
search s and f r libr y design or com-
pound lection for purchasing cam-
paigns of ta get class focused com-
pou d collections. Linking the leaf
nodes of the ligand-target classification
tree to the sequences accession codes
(e.g., SWISS-PROT AC) of the precise
molecu r t rgets allows BLAST-type
sequ nce similarity-based identifica-
tion of the li ands of the next homolo-
gous rece t rs; ligands in this sense
relating to sequences.
The ligand-target classification can
also be used for the ana ysis of corpo-
ate high-throughput screening and
profiling data, which are a very rich
source of structure-activi y data,
including large amounts f proprietary
data. If one links each assay to a target
no e i the class fication scheme, it is
possible to select all assays related to a
targ t family. In a second step, all com-
pounds that showed activity in at least
one f them can be used to collect the
ompounds active for a target family.
These compounds can then be submit-
t d o ssays related targets or serve
as reference structur s for further i il-
icoscree ng or design of target class
focused libraries. Both disciplines rely
on the possibility to retrieve compre-
hensive ets of ligands that are likely to
share a conserved molec lar recogni-
tion mode. Within the frame of the set-
p of n integrated platform for the
analys s of HTS and profiling data, we
are thu currently workin on the
implemen ation of corresponding
anno ation schemes for HTS and pro-
filing data.
94 Drug News Perspect 16(2), March 2003