384

Chemical space navigation in lead discovery

Tudor I Oprea
The number of new chemical entities has remained rather biological data, through combinatorial chemistry [2••], and
constant (averaging 37 per year) in the past decade, despite high-throughput screening (HTS) [3]. However, this
the multiple-fold increase in the number of compounds that are plethora of chemical and biological data has yet to translate
being made and tested. Chemical space requires novel into clinical success [4•], despite the logical progression of
methods that can handle the increasing number of potentially HTS actives into lead compounds and drug candidates to
accessible molecules. Neighborhood behavior, as an approach the launched status [5•]. In fact, the number of new
to similarity, and chemical property space navigation are some chemical entities (NCEs) has remained rather constant in
of the recent advances that are discussed, in the context of the past decade [6•], even though the number of molecules
lead discovery and appropriate pharmacokinetic properties. being synthesized and screened increased by several
orders of magnitude during the same period. Because poor
Addresses pharmacokinetic behavior is, besides poor efficacy, the
EST Lead Informatics, AstraZeneca R&D Mölndal, S-43183 Mölndal, main reason for failure in the clinical trials [7], this review
Sweden; e-mail: [email protected] focuses on chemical space navigation tools, in particular on
Current Opinion in Chemical Biology 2002, 6:384–389 methods that increase the awareness for pharmacokinetic
properties, besides molecular diversity.
1367-5931/02/$ — see front matter
© 2002 Elsevier Science Ltd. All rights reserved.
Chemical similarity validation
Published online 15 April 2002 One of the misleading concepts in chemical space navigation
Abbreviations
relates to chemical, or molecular similarity [8,9,10••–12••].
FBPF fuzzy bipolar pharmacophore fingerprint Similarity is, by definition (Box 1), related to a particular
HTS high-throughput screening framework: that of a descriptor system (we need a metric
NB neighborhood behavior by which to judge similarity), as well as that of an object,
NCE new chemical entity
or class of objects (we need a reference point to which
PCA principal component analysis
objects can be compared). Similarity depends on the
choice of molecular descriptors [13•], the choice of the
Introduction weighting scheme(s) and the similarity coefficient itself [10••].
The vastness of chemical space cannot be better illustrated Its essence, thus, remains in the eye of the beholder: their
than by using David Weininger’s example [1]: considering nature will classify oranges and tomatoes in a different
all the derivatives of n-hexane, starting from a list of 150 category (i.e. fruit vs. vegetable), although they appear to
substituents, and enumerating all the possibilities, from be quite similar from a geometrical perspective (spheroidal,
mono- to 14-substituted hexanes, one can easily reach over perhaps 7–15 cm in diameter). In fact, even in the macro
1029 structures. While most of them might be, to date, world, similarity perception is not as trivial as one might
synthetically inaccessible, a small number of building think, as illustrated by the fruit vs. vegetable debate — see
blocks can still give an infinite number of possibilities, as Box 2. By the same token, 3D-based pharmacophore [14]
seen in living systems: 20 natural amino acids and five fingerprints [15] are expected to afford different similarity
nucleotides form the basis of most proteins and nucleic results than, for example, the Daylight fingerprints [16],
acids, respectively. As chemical space is limited at the as recently shown by Dixon and Merz [17•], but not
low-molecular-weight end only, methods to navigate it are necessarily because one metric is better than the other.
becoming increasingly important. Not only for molecular Rather, one is expected to use different similarity or,
biology and material science, but also for medicinal indeed, diversity metrics for different purposes.
chemistry — and in particular for lead discovery.
For example, ‘supine’ is dissimilar when compared to
The pharmaceutical industry has benefited, in recent ‘ovine’, ‘lupine’ and ‘vulpine’ — which, in turn, can be
years, from a technological breakthrough in terms of rapid clustered (i.e. similar) when the choice is from this list
access to a large number of novel compounds and related alone, but all of the above would stand out as outliers
(i.e. dissimilar) in a comprehensive list of cities in Sweden.
Thus, the misleading character of ‘molecular (dis)similarity’
Box 1
relates to our (in)ability to suitably compare the proverbial
According to the Webster’s New World Dictionary & Thesaurus apples and oranges, while choosing a metric that is most
(Macmillan Publishers), sim·i·lar·i·ty (VLP′ʴOHU′ʴWƝ), n., pl. -ties, is 1.
fitting for our requirements. If one is tempted to estimate
the state or quality of being similar; resemblance or likeness; 2. a
point, feature, or instance in which things are similar. Synonyms for the validity of any molecular (dis)similarity metric by
similarity include: likeness, connection, comparison, parallel, comparing its output to the binding affinities of ligands to
correspondence, match, and relationship — all of which indicate that a promiscuous target like albumin or the 3A4 isozyme of
a point of reference is required.
cytochrome P450, one should not neglect the fact that such
 Chemical space navigation in lead discovery Oprea 385

targets have a built-in ability to bind an impressive variety Box 2

of endogenous and exogenous structures. On the other “Botanically speaking, the tomato is a fruit, but legally speaking it is
hand, biological targets that are much more precise in a vegetable. The Supreme Court of the United States said so in
terms of the molecular recognition features that are likely 1893. An importer had argued that tomatoes were fruit and
to result in high-affinity, high-specificity ligands, are more therefore not subject to a duty in effect at that time. The Court held
that the tomato is a vegetable because it was usually served at
likely to assist us in validating and refining chemical dinner in, with, or after the soup, or with fish or meats that constitute
similarity methods that, in turn, can be used to rapidly scan the main part of the meal. This is less true now than it was then, for
large databases for ligands of interest. The higher the today a much larger part of our tomato crop is made into juice, but
specificity and selectivity of our target, the more likely we the tomato remains, legally, a vegetable.”
are to find that molecular similarity tools become relevant. Extracted from http://aggie-horticulture.tamu.edu/
By the same token, the term ‘maximum diversity’ can not, plantanswers/vegetables/vegancestors.html
or should not, be taken out of context: maximum diversity
can only refer to the number of combinations available in a two sets of the same size (one random, one NB-selected).
limited pool of reagents [18–20] but should be regarded as The FBPF-based autocorrelograms [29] have been further
an inefficient strategy for chemical space navigation [21••]. benchmarked against other similarity metrics (Horvath D,
Chemical diversity is infinite — our understanding of it is Jeandenans C, personal communication), and found to be
only limited by time and resources. significantly more appropriate in capturing neighborhood
behavior in a more general manner.
Chemical space neighborhood and lead-like
space In a different study, the autocorrelogram-based ALMOND
When compounds need to be prioritized for synthesis, or descriptors [31] (ALMOND is available from Multivariate
even for HTS assays [22•], one needs to ask the question: Infometric Analysis srl, Perugia, Italy, http://www.miasrl.com)
serendipity [23••] notwithstanding, how similar (or how have been shown to provide reliable quantitative
diverse) do I want my compound library to be, based on structure–activity models for a range of targets: glycogen
prior target-related knowledge? Cheminformatics [24,25•] phosphorylase b, corticosteroid-binding globulin and
has an increasingly important role in the decision-making serotoninergic (5-HT2A) receptors, thus providing a novel
process [26], namely in constantly re-evaluating biologically class of alignment-independent, chemically interpretable,
relevant information, both at the chemical (2D, 3D) and pharmacophore-related descriptors. One can expect
structural (3D) level, in the lead and drug discovery autocorrelogram-based pharmacophore tools to emerge as
process (see Figure 1). Reagent selection is more than ever a method of choice in exploring neighborhood behavior,
dependent on chemical space navigation tools, and should and probing chemical space.
be guided by structural chemistry and biological testing: as
the information content increases, one should use targeted The NB concept is a similarity-rooted extension of the
diversity methods [27,28] in order to maximize the chances related medicinal chemistry tenet of ‘lead series’: struc-
of finding active compounds. turally related chemicals (lead series) are expected to
exhibit a good range of biological activities (preferably
The task of navigating in chemical space is more challenging three orders of magnitude or more), to provide the basis for
in the absence of 3D-structural target information. An (quantitative) structure–activity relationship analyses. Besides
approach that goes beyond ‘maximum diversity’ starts the good pharmacokinetic properties required for high-
from a pre-defined activity space for over 584 drugs and quality leads (discussed in the next section), other features
drug-like compounds, tested on 42 different targets of promising lead series include favorable intellectual
(Horvath D, Jeandenans C, personal communication), property position and relatively simple structures, suitable
activity space that allows an unambiguous mapping in for further chemical alterations of chosen molecules
chemical and biological space. Similarity is then validated (‘leads’) in order to optimize the target activity. The ability
against two criteria: ‘completeness’ (i.e. the number of to yield such lead series from combinatorial libraries has
compound pairs with similar activity profiles that have been put to question [32•], based on Lipinski’s earlier work
high similarity scores), and ‘consistency’ (i.e. the number [33] that analyzed high-molecular-weight, low-polarity
of compound pairs with different in vitro profiles that have trends at Pfizer. This led to in-depth analyses regarding
low similarity values) (Horvath D, Jeandenans C, personal the exact nature of chemical leads [34,35] that established
communication). These two criteria are used to validate a the presence of a lead-like space. Thus, on average, leads
novel similarity metric, based on fuzzy bipolar pharmacophore are expected to have lower molecular complexity [34]
fingerprints (FBPFs) [29], that relies on the concept of when compared with drugs (fewer rings, fewer rotatable
neighborhood behavior (NB) [30]. An extension of the bonds), to be smaller (lower molecular weight), and more
similarity postulate, which states that structurally similar polar [35]. Furthermore, the intellectual property position
molecules are expected to display similar biological activity, in a lead series could be strengthened by theoretical or
the NB concept states that more active compounds than experimental means. On the theoretical side, lead hopping
random are expected to be found in a subset of the [36] can be used by jumping across different series [37]
structurally nearest neighbors of an active compound, given using the 3D-based topomer similarity technique [38]
386 Combinatorial chemistry

Figure 1

The chemical information flow in the drug-

Structural / computational chemistry discovery process: chemical space navigation
allows one to select reagents
(e.g. (a) heterocyclic scaffolds, (b) amino
+ O
acids and (c) spacers) that yield, via multiple
N N N
O N parallel syntheses, (d) products, which are
CI
N N N N N N Structural biologically tested by HTS. Structural
O information information (from X-ray, NMR or
(e) (f) computational chemistry models) could then
provide an in-depth understanding of the
intermolecular forces that are relevant for the
biological target in question (e.g. (e) steric,
electrostatic and π–π interactions), which may
Combinatorial chemistry / HTS O lead to (f) optimized ligands. Red arrows
Biological
indicate a logical progression, whereas white
N N O information
N O arrows indicate the interdependence between
O N O chemical, biological and structural information.
N N N N N
O O O O N O Line arrows indicate feedback loops.
(d)

Chemical space navigation Chemical

information
N N N N
N O
N O N
N S
O N O

O N
N
N N O O
(a) (b) (c)

Current Opinion in Chemical Biology

(Topomer-based similarity techniques are available from space mapping effort requires a property-based system,
Tripos Inc., St Louis, Missouri, http://www.tripos.com), or besides an efficient similarity/diversity metric.
by surfing across the ‘scaffold space’ with SORT&gen [39]
(SORT&gen is available from SPECS and BioSPECS, Early attempts to map physical properties described the
Rijswijk, The Netherlands, http://www.specs.net). On the two-dimensional BC(DEF), ‘bulk’ and ‘cohesiveness’
experimental side, one could resort to high-energy gamma- parameters [42], derived from six physical properties
ray radiations (Kessler U, Pilger BD, Zerbe O, Scapozza L, (aqueous solvation energy, partition coefficient, boiling
Folkers G, personal communication) or to high-speed point, molecular refractivity, volume and vaporization
microwave chemistry [40]. enthalpy) for a set of 114 pure liquids. This scheme was
shown to work quite well for a set of 139 diverse structures
Chemical property space navigation [43]. By analogy to the Mercator convention in geography,
Using the analogy of an intercity distances table, in contrast we recently suggested chemography, a combination of
to a geographical map, Martin and Critchlow [41•] pointed rules (not unlike the BC(DEF) dimensions) and objects
out the advantage of having a chemical space map, rather (chemical structures), to provide a consistent, global
than mere distance-based ‘diversity’ in combinatorial chemical space map [44,45]. Chemographic rules included,
library design. Having the right inter-object distances is initially, general properties such as size, lipophilicity, and
clearly not enough, as one is likely to be successful in hydrogen bond capacity, while objects include ‘satellites’,
finding a list of, for example, five cities in Western intentionally placed outside the drug-like space, as well as
Europe that have identical (or close) distances to cities on ‘core’ objects, selected mostly from a list of orally available
the East Coast of the United States. In the absence of a drugs. ChemGPS, the chemical global positioning system,
proper map, a sixth city in Eastern Europe could, in the comprises both the ‘core’ and ‘satellite’ molecules.
wrong context, be placed somewhere in the Atlantic Chemographic map coordinates are extracted, in ChemGPS,
Ocean. In other words, context-sensitive information is by principal component analysis (PCA) [46], from a (fixed)
required while evaluating chemical spaces, even though list of molecular descriptors that evaluate the above-
appropriate measures may have been taken with respect mentioned rules on a single set of molecules. PCA-score
to distance-based (dis)similarity. Thus, an effective chemical prediction is used, then, to project new molecules on the
 Chemical space navigation in lead discovery Oprea 387

pre-established map — providing thus a consistent and Figure 2

systematic method to map the chemical property space. A
map derived with the same choice of rules (dimensions)
Large
and the same set of compounds (objects) can then be used
Hydrophobic
to compare PCA scores across a large number of chemicals, Rigid
Small
because it does not change with chemistry and time. Hydrophobic
ChemGPS is therefore well suited to become a reference Rigid Large
So Hydrophobic
system for comparing multiple combinatorial libraries, and lub Limited flexibility
ility
for keeping track of previously explored regions of the
Small ility
chemical space. eab
Hydrophyllic Perm
Rigid ?
The ability of ChemGPS to provide a global property
space map was evaluated using atom-count, atom-type
count, topological, electrostatic and physico-chemical Small Large
property descriptors [44] derived from 2D-structures, for a Hydrophyllic Polar
Flexible Limited flexibility
set of 45 heteroaromatic compounds [47] previously
described with GRID [48], for a set of 87 α-amino acids Current Opinion in Chemical Biology

with known z-scores (principal properties) [49], as well as

for a set of 8600 monocarboxylic acids, for which extensive
Chemical space map interpretation based on the combined ChemGPS
comparisons between local and global PCA models models derived from 2D (black arrows) and 3D (colored arrows)
were performed. Based on the above descriptors, a nine- descriptors. Small structures are located on the left side, whereas large
dimensional map was derived that could be interpreted in structures are on the right side. Hydrophobic compounds occur more
terms of, for example, size, hydrophobicity and flexibility often on the upper side, and hydrophilic compounds are more frequent
on the lower side. Rigid structures are more frequent below the paper
(see Figure 2). plane, whereas flexible structures occur more often above. Permeability
has some degree of correlation with size and hydrophobicity, whereas
Pharmacokinetic property prediction is currently being solubility cannot be rigorously placed.
considered early on in the process of drug discovery — to
the extent that combinatorial chemistry libraries are being
scrutinized for ADME (absorption, distribution, metabo- Conclusions
lism and excretion) properties [50,51]. This has recently Recent advances in the area of neighborhood behavior
been extended to virtual screening efforts [52,53]. and chemical property space mapping enable us to
GPSVS, the combination of ChemGPS molecules complement ‘chemical similarity’ in ways that are mean-
(objects) and VolSurf descriptors (Volsurf is available from ingful for combining chemical, biological and structural
Molecular Discovery Ltd, London, UK, http://www.mold- information. Pharmacokinetic and chemical property space
iscovery.com) [54] (rules) was used to map the chemical can be simultaneously queried, together with biological
space with respect to passive permeability and activity space and chemical fingerprints or pharmacophores.
solubility — key properties according to the US Food and This is likely to enable chemistry projects to reach an
Drug Administration [55]. VolSurf has been extensively informed decision in the early stages of lead identification,
validated in both oral absorption [56,57] and blood-brain- thus allowing the progression of higher-quality leads
barrier permeation [58•] models. We have recently shown through the drug-discovery pipeline.
[59] that, in the absence of any biological input, GPSVS
(a model based on PCA prediction), correlates well with Acknowledgements
I would like to thank Drs Katherine Andrews Cramer and Richard D Cramer
passive transcellular permeability, as illustrated for the (both from Tripos, Inc.) and Dr Dragos Horvath (from CEREP) for sharing
Caco-2, ghost erythrocyte and blood-brain barrier datasets their work prior to publication.
— for the first dimension, and with solubility, as investigated
using the octanol–water partition, and intrinsic solubility References and recommended reading
Papers of particular interest, published within the annual period of review,
datasets — for the second dimension, respectively. The have been highlighted as:
two GPSVS dimensions are no longer orthogonal, forming • of special interest
an angle of approximately 43° — see also Figure 2. The •• of outstanding interest
global property map of chemical space, starting from 1. Weininger D: Combinatorics of small molecular structures. In
2D-based [44] and 3D-based [59] descriptors is depicted Encyclopedia of Computational Chemistry, Vol 1. Edited by
Von Ragué Schleyer P, Allinger NL, Clark T, Gasteiger J, Kollman PA,
in Figure 2: a general sense of property location can be Schaefer HF III. New York: Wiley; 1998:425-530.
derived, based on the fact that the first dimension in 2. Lebl M: Parallel personal comments on ‘classical’ papers in
the two ChemGPS models has a weak (r2 = 0.61) corre- •• combinatorial chemistry. J Comb Chem 1999, 1:3-24.
lation. The question mark in the solubility arrow is a Insightful, personal comments and key references from authors involved
early on in the field of combinatorial chemistry.
reminder that the exact relationship between chemical
3. Fox S, Farr-Jones S, Yund MA: High throughput screening for drug
properties, in a general sense, and aqueous solubility, discovery: continually transitioning into new technology. J Biomol
needs further investigation. Screening 1999, 4:183-186.
388 Combinatorial chemistry

4. Horrobin DF: Innovation in the pharmaceutical industry. J Royal 23. Kubinyi H: Chance favors the prepared mind — from serendipity
• Soc Med 2000, 93:341-345. •• to rational drug design. J Rec Signal Transduction Res 1999,
This paper articulates some of the issues related to marketing-driven 19:15-39.
preclinical research in the pharmaceutical industry. Fast-paced review of serendipitous drug discoveries, emphasizing structure-
based design aspects. “Screening, especially…HTS, can be considered as
5. Drews J: Drug discovery: a historical perspective. Science 2000, a systematic approach to benefit from mere chance”.
• 287:1960-1964.
A synopsis of the drug-discovery process is provided. 24. Brown F: Chemoinformatics: what is it and how does it impact
drug discovery. Annu Rep Med Chem 1998, 33:375-384.
6. Gaudillière B, Bernardelli P, Berna P: To market, to market — 2000.
• Annu Rep Med Chem 2000, 36:293-318. 25. Hahn MM, Green R: Cheminformatics — a new name for an old
Each year, Annual Reports in Medicinal Chemistry summarizes the NCEs • problem? Curr Opin Chem Biol 1999, 3:379-383.
introduced into the world market for the first time in the previous year. 35 Both software and hardware are discussed from a practitioner’s perspective.
NCEs were introduced into their first markets in 2000. Previous years are as
follows, with the number of NCEs first introduced in the market that year 26. Olsson T, Oprea TI: Cheminformatics: a tool for decision-makers in
given in brackets: 1991 (36), 1992 (36), 1993 (43), 1994 (a record 44 NCEs), drug discovery. Curr Opin Drug Discov Dev 2001, 4:308-313.
1995 (35), 1996 (38), 1997 (39), 1998 (27), and 1999 (35). The average
27. Li J, Murray CW, Waszkowycz B, Young SC: Targeted molecular
for this decade is 37 NCEs.
diversity in drug discovery — integration of structure-based design
7. Kennedy T: Managing the drug discovery/development interface. and combinatorial chemistry. Drug Discov Today 1998, 3:105-112.
Drug Discov Today 1997, 2:436-444.
28. Lewis RA: The design of small- and medium-sized focused
8. Willett P: Similarity and Clustering Techniques in Chemical combinatorial libraries. In Molecular Diversity in Drug Design.
Information Systems. Letchworth: Research Studies Press; 1987. Edited by Dean PM, Lewis RA. Dordrecht: Kluwer Academic
Publishers; 1999:221-248.
9. Johnson MA, Maggiora GM: Concepts and Applications of Molecular
Similarity. New York: Wiley; 1990. 29. Horvath D: High throughput conformational sampling and fuzzy
similarity metrics: a novel approach to similarity searching and
10. Willett P: Chemoinformatics — similarity and diversity in chemical focused combinatorial library design and its role in the drug
•• libraries. Curr Opin Biotechnol 2000, 11:85-88. discovery laboratory. In Combinatorial Library Design and
Good overview of cluster-based, partition-based, dissimilarity-based and Evaluation for Drug Design. Edited by Ghose AK, Viswanadhan VN.
optimization-based selection techniques for combinatorial synthesis planning. New York: Marcel Dekker Inc.; 2001:429-472.
11. Lewis RA, Pickett SD, Clark DE: Computer-aided molecular 30. Patterson DE, Cramer RD, Ferguson AM, Clark RD, Weinberger LE:
•• diversity analysis and combinatorial library design. Rev Comput Neighborhood behavior: a useful concept for validation of
Chem 2000, 16:1-51. ‘molecular diversity’ descriptors. J Med Chem 1996, 39:3049-3059.
This is an exhaustive review related to molecular similarity and diversity.
Methods to probe chemical diversity, including molecular descriptor validation, 31. Pastor M, Cruciani G, McLay I, Pickett S, Clementi S:
are discussed in the context of combinatorial library design. GRid-INdependent Descriptors (GRIND): a novel class of
alignment-independent three-dimensional molecular descriptors.
12. Martin YC: Diverse viewpoints on computational aspects of J Med Chem 2000, 43:3233-3243.
•• molecular diversity. J Comb Chem 2001, 3:231-250.
This review provides first-hand accounts and key references from authors 32. Teague SJ, Davis AM, Leeson PD, Oprea TI: The design of leadlike
involved early on in the field of molecular similarity and diversity. • combinatorial libraries. Angew Chem Int Ed Engl 1999,
38:3743-3748. German version: Angew Chem 1999, 111:3962-3967.
13. Todeschini R, Consonni V: Handbook of Molecular Descriptors. Low-molecular weight and low-hydrophobicity are discussed as key properties
• Weinheim: Wiley-VCH; 2000. when designing combinatorial libraries.
This book covers ca. 3000 molecular descriptors.
33. Lipinski CA, Lombardo F, Dominy BW, Feeney PJ: Experimental and
14. Gund P: Three-dimensional pharmacophoric pattern searching. computational approaches to estimate solubility and permeability
Prog Mol Subcell Biol 1977, 11:117-143. in drug discovery and development settings. Adv Drug Deliv Rev
1997, 23:3-25.
15. Olender R, Rosenfeld R: A fast algorithm for searching for
molecules containing a pharmacophore in very large virtual 34. Hann MM, Leach AR, Harper G: Molecular complexity and its
combinatorial libraries. J Chem Inf Comput Sci 2001, 41:731-738. impact on the probability of finding leads for drug discovery.
J Chem Inf Comput Sci 2001, 41:856-864.
16. Daylight Chemical Information System, Santa Fe, New Mexico.
http://www.daylight.com 35. Oprea TI, Davis AM, Teague SJ, Leeson PD: Is there a difference
between leads and drugs? A historical perspective. J Chem Inf
17. Dixon SL, Merz KM: One-dimensional molecular representations
Comput Sci 2001, 41:1308-1315.
• and similarity calculations: methodology and validation. J Med
Chem 2001, 44:3795-3809. 36. Cramer RD, Poss MA, Hermsmeier MA, Caulfield TJ, Kowala MC,
This paper provides a simplified, one-dimensional projection of molecular Valentine MT: Prospective identification of biologically active
structures, that can be generated from 2D or 3D structures. structures by topomer shape similarity searching. J Med Chem
1999, 42:3919-3933.
18. Leach AR, Bradshaw J, Green DVS, Hann MM: Implementation of a
system for reagent selection and library enumeration, profiling, 37. Cramer RD, Jilek RJ, Andrews KM: Dbtop: topomer similarity
and design. J Chem Inf Comput Sci 1999, 39:1161-1172. searching of conventional structure databases. J Mol Graph Model
2002, in press.
19. Shi S, Peng Z, Kostrowicki J, Paderes G, Kuki A: Efficient
combinatorial filtering for desired molecular properties of reaction 38. Andrews KM, Cramer RD: Toward general methods of targeted
products. J Mol Graph Model 2000, 18:478-496. library design: topomer shape similarity searching with diverse
structures as queries. J Med Chem 2000, 43:1723-1740.
20. Lobanov VS, Agrafiotis DK: Stochastic similarity selections from
large combinatorial libraries. J Chem Inf Comput Sci 2000, 39. De Laet A, Hehenkamp JJJ, Wife RL: Finding drug candidates in
40:460-470. virtual and lost/emerging chemistry. J Heterocyclic Chem 2000,
21. Lipinski CA: Drug-like properties and the causes of poor solubility 37:669-674.
•• and poor permeability. J Pharmacol Toxicol Methods 2000, 40. Kappe CO: High-speed combinatorial synthesis utilizing
44:235-249. microwave irradiation. Curr Opin Chem Biol 2002, 6:this issue.
Analysis of clinical candidates at Merck indicates that the ‘rational design
approach’ leads to poorer permeability, whereas Pfizer’s ‘HTS approach’ 41. Martin EJ, Critchlow RE: Beyond mere diversity: tailoring
appears to result in clinical candidates with poorer solubility. • combinatorial libraries for drug discovery. J Comb Chem 1999,
1:32-45.
22. Agrafiotis DK, Rassokhin DN: Design and prioritization of plates Chemical property binning, via multi-dimensional scaling, could facilitate
• for high-throughput screening. J Chem Inf Comput Sci 2001, chemical diversity void identification.
41:798-805.
The issue of plate selection for HTS is discussed in view of plate-based 42. Cramer RD: BC(DEF) parameters: 1. The intrinsic dimensionality
molecular diversity and similarity to known lead(s). User-defined selection of intermolecular interactions in the liquid state. J Am Chem Soc
criteria (e.g. purity), can also be included in these experimental design schemes. 1980, 102:1837-1849.
 Chemical space navigation in lead discovery Oprea 389

43. Cramer RD: BC(DEF) parameters: 2. An empirical structure-based 52. Oprea TI, Zamora I, Svensson P: Quo vadis, scoring functions?
scheme for the prediction of some physical properties. J Am Toward an integrated pharmacokinetic and binding affinity
Chem Soc 1980, 102:1849-1859. prediction framework. In Combinatorial Library Design and
Evaluation for Drug Design. Edited by Ghose AK, Viswanadhan VN.
44. Oprea TI, Gottfries J: Chemography: the art of chemical space New York: Marcel Dekker Inc; 2001:233-266.
navigation. J Comb Chem 2001, 3:157-166.
53. Oprea TI: Virtual screening in lead discovery: a viewpoint.
45. Oprea TI, Gottfries J: ChemGPS: a chemical space navigation tool. Molecules 2002, 7:51-62.
In Rational Approaches to Drug Design. Edited by Höltje HD,
Sippl W. Barcelona: Prous Science Press; 2001:437-446. 54. Cruciani G, Crivori P, Carrupt PA, Testa B: Molecular fields in
quantitative structure-permeation relationships: the VolSurf
46. Jackson JE: A Users Guide to Principal Components. New York: approach. J Mol Struct (Theochem) 2000, 503:17-30.
Wiley; 1991.
55. Anonymous: Waiver of in vivo bioavailability and bioequivalence
47. Clementi S, Cruciani G, Fifi P, Riganelli D, Valigi R, Musumarra G: studies for immediate-release solid oral dosage forms based on
A new set of principal properties for heteroaromatics obtained by a biopharmaceutics classification system. 2000. Available from
GRID. Quant Struct Act Relat 1996, 15:108-120. http://www.fda.gov/cder/OPS/ BCS_guidance.htm.
48. Goodford PJ: Computational procedure for determining 56. Guba W, Cruciani G: Molecular field-derived descriptors for the
energetically favourable binding sites on biologically important multivariate modeling of pharmacokinetic data. In Molecular Modeling
macromolecules. J Med Chem 1985, 28:849-857. and Prediction of Bioactivity. Edited by Gundertofte K, Jørgensen FS.
New York: Kluwer Academic/Plenum Publishers; 2000:89-94.
49. Sandberg M, Eriksson L, Jonsson J, Sjöström M, Wold S: New
chemical descriptors relevant for the design of biologically active 57. Zamora I, Oprea TI, Ungell AL: Prediction of oral drug permeability.
peptides. A multivariate characterization of 87 amino acids. J Med In Rational Approaches to Drug Design, Edited by Höltje HD,
Chem 1998, 41:2481-2491. Sippl W. Barcelona: Prous Science Press; 2001:271-280.
50. Darvas F, Dorman G: Early integration of ADME/Tox parameters 58. Crivori P, Cruciani G, Carrupt PA, Testa B: Predicting blood-brain
into the design process of combinatorial libraries. Chim Oggi • barrier permeation from three-dimensional molecular structure.
1999, 17:10-13. J Med Chem 2000, 43:2204-2216.
A comprehensive list of drugs that penetrate (or not) the blood-brain-barrier.
51. Pickett SD, McLay IM, Clark DE: Enhancing the hit-to-lead
properties of lead optimization libraries. J Chem Inf Comput Sci 59. Oprea TI, Zamora I, Ungell AL: A pharmacokinetically based mapping
2000, 40:263-272. device for chemical space navigation. J Comb Chem 2002, in press.