Faulon J.L., Bender A. Handbook of Chemoinformatics Algorithms
Подождите немного. Документ загружается.
288 Handbook of Chemoinformatics Algorithms
ways to match the target molecular Signature. More details on implementation and
examples are provided elsewhere [19].
Once the structures are generated, postprocessing of these candidates can occur to
remove those structures that are not reasonable (such as multiple bridges or aromatic
rings not following Huckel’s rule). Additional filtering can occur based on energy
minimization via the use of force fields, if desired.
ALGORITHM 9.10 INVERSE DESIGN WITH SIGNATURE: STRUCTURE
GENERATION AND ANALYSIS
1. Submit molecular signatures to structure generation code
2. Evaluate structures based on stability/energetic constraints
a. Use of Huckel’s Rule
b. Use of intramolecular force field
At this stage, the algorithm is complete and the resulting structures that have passed
through postprocessing steps form a focused database of candidate solutions for the
problem at hand. Here, these structures may be bought (if available) or synthesized and
then evaluated by other means (such as experimentation) in order to gain additional
confidence in the solutions.
9.5.2.3 Case Study: Chemical Process Industry Application
Faulon and coworkers have implemented the above algorithm to design novel poly-
mers with desired properties [5]. They used 33 polymers and focused on glass
transition temperature, heat capacity, and density. Height-1 atomic signatures were
used on the repeat unit of the polymer. This resulted in 63 unique height-1 atomic
signatures and, accordingly, 28 constraint equations. Instead of using the brute force
approach to solve the constraint equations, they employed an implementation of the
Fortenbacher algorithm [64] to arrive at basis vectors for their solutions. From there
they calculated linear combinations of the basis vectors and arrived at over 800 million
molecular Signatures (with the constraint that no repeat unit could have more than
50 atoms). These solutions were then filtered on a second constraint that no solution
could have an occurrence number for a particular atomic Signature above that in the
original dataset. Additionally, a normalized Euclidean distance metric was used so
that solutions only in the nearest neighborhood of the original dataset were used. Of
the more than 800 million solutions exposed to these constraints, only 1327 were
acceptable.
When the 1327 molecular signatures were exposed to the three property constraints
(313 K for glass transition temperature, 439 J/mol-K for heat capacity, 1.04 g/mol
for density; all plus or minus an absolute mean error), 80 solutions were left. From
this list, 178 structures were generated. A small number of the structures with some
of the predicted target properties are shown in Figure 9.7.
9.5.2.4 Case Study: Bio-Related Application
Recently, Visco and coworkers used the inverse design algorithm with Signature to
design novel glucocorticoid receptor ligands with pulmonary selectivity [65]. They
Computer-Aided Molecular Design 289
O O
O
N
O
O
N
O
T
g
= 294.1 K
C
p
= 426.3 J/mol-
K
ρ = 1.04 g/cm
3
T
g
= 299.9 K
C
p
= 437.8 J/mol-
K
ρ = 1.06 g/cm
3
FIGURE 9.7 Two of the 137 polymer repeat units that were identified as having predicted
targeted properties by the inverse design algorithm using the Signature molecular descriptor.
O
HO
O
O
Cl
O
O
FIGURE 9.8 A compound predicted to be an effective glucocorticoid receptor ligand using
the inverse design technique with the Signature molecular descriptor.
explored properties such as high receptor binding affinity, high systemic clearance,
high plasma protein binding, and low oral bioavailability using atomic height-2 Signa-
tures designed from previously published data to identify 84 high priority compounds
predicted to be at least or more effective than currently available corticosteroids. One
such compound developed, which is nonsteroidal in nature, is provided in Figure 9.8.
The optimal structure had a predicted oral bioavailability of 1%, a plasma protein
binding of 99.1%, a systemic clearance of 475 L/h, and a relative receptor binding
affinity of 1179.45.
9.6 CONCLUDING REMARKS
In this chapter, we examined three CAMD algorithms that are being implemented
to solve molecular design problems of both an engineering-related and a biological
290 Handbook of Chemoinformatics Algorithms
nature. Each of the techniques has its own strengths and weaknesses. For the Hybrid-
CAMD approach, while it does combine information with increasing complexity
to mitigate some of the combinatorial explosion, one is still limited (at least in the
lowest levels) to the groups selected at the start of the algorithm. Additionally, to
design larger molecules, templates are seemingly necessary in order to focus the
solution space to an area where higher-quality solutions are expected. For the CAMD
as optimization approach, high-quality optimization codes can be brought to bear on a
problem formation, thus providing an access to these techniques. However, templating
is required to make progress on problems that limit the solution space. Finally, for
the CAMD approach with Signature, one is not limited by an initial set of fragments
and templating is not required, which can open up nonintuitive areas of the chemical
space. However, storage issues and computational time issues can quickly become
bottlenecks when the initial dataset molecules become large.
For all three of these techniques, a combination of increased computational and
storage resources combined with algorithm refinement will make them viable CAMD
techniques for the near future.
REFERENCES
1. Brown, N. and R.A. Lewis, Exploiting QSAR methods in lead optimization. Curr. Opin.
Drug Discov. Devel., 2006, 9(4): 419–424.
2. Clark, D.E., What has computer-aided molecular design ever done for drug discovery?
Expert Opin. Drug Discov., 2006, 1(2): 103–110.
3. de Julian-Ortiz, J.V., Virtual Darwinian drug design: QSAR inverse problem, virtual com-
binatorial chemistry and computational screening. Comb. Chem. High Throughput Screen.,
2001, 4(3): 295–310.
4. Taft, C.A., V.B. Da Silva, and C. Da Silva, Current topics in computer-aided drug design.
J. Pharma Sci., 2008, 97(3): 1089–1098.
5. Brown,W.M., et al., Designing novel polymers with targeted properties using the signature
molecular descriptor. J. Chem. Inf. Model., 2006, 46: 826–835.
6. Camarda, K.V. and C.D. Maranas, Optimization in polymer design using connectivity.
Ind. Eng. Chem. Res., 1999, 38: 1884–1892.
7. Gani, R., et al., A modern approach to solvent selection. Chem. Eng., 2006, March:
30–43.
8. Dobson, C.M., Chemical space and biology. Nature, 2004, 432: 824–828.
9. Garg, S. and L.E.K. Achenie, Mathematical programming assisted drug design for
nonclassical antifolates. Biotechnol. Prog., 2001, 17: 412–418.
10. Lin, B., et al., Computer-aided molecular design using tabu search. Comput. Chem. Eng.,
2005, 29: 337–347.
11. Schneider, G. and U. Fechner, Computer-based de novo design of drug-like molecules.
Nat. Rev. Drug Discov., 2005, 4: 649–663.
12. Todorov, N.P., I.L. Alberts, and P.M. Dean, De novo design. In: Comprehensive Medicinal
Chemistry II, J.B. Taylor and D.J. Triggle (eds). Elsevier Ltd.: Oxford, pp. 283–305, 2007.
13. Brignole, E.A., S. Bottini, and R. Gani, A strategy for design and selection of solvents for
separation processes. Fluid Phase Equi., 1986, 29: 125–132.
14. Gani, R. and E.A. Brignole, Molecular design of solvents for liquid extraction based on
UNIFAC. Fluid Phase Equi., 1983, 13: 331–340.
Computer-Aided Molecular Design 291
15. Gani, R., Case studies in chemical product design—use of CAMD techniques. In: Chemi-
cal Product Design: Towards a Perspective Through Case Studies, 23, K.M. Ng, R. Gani,
and K. Dam-Johansen (eds). Elsevier: Amsterdam, the Netherlands, pp. 435–458, 2007.
16. Raman, V.S. and C.D. Maranas, Optimization in product design with properties correlated
with topological indices. Comput. Chem. Eng., 1998, 22: 747–763.
17. Camarda, K. and P. Sunderesan, An optimization approach to the design of value-added
soybean oil products. Ind. Eng. Chem. Res., 2005, 44: 4361–4367.
18. Churchwell, C., et al., The signature molecular descriptor. 3. Inverse quantitative structure
relationship of ICAM-1 inhibitory peptides. J. Mol. Graph. Model., 2003, 22: 263–273.
19. Faulon, J.-L., C.J. Churchwell, and D.P. Visco Jr., The signature molecular descriptor. 2.
enumerating molecules from their extended valence sequences. J. Chem. Info. Comput.
Sci., 2003, 43: 721–734.
20. Hansch, C., et al., Correlation of biological activity of phenoxyacetic acids with Hammett
substituent constants and partition coefficients. Nature, 1962, 194: 178–180.
21. Gong, Z., et al., Quantitative structure–activity relationship study on fish toxicity of
substituted benzenes. QSAR Comb. Sci., 2008, 27: 967–976.
22. Selassie, C.D., History of quantitative structure–activity relationships. In: Burger’s Medic-
inal Chemistry and Drug Discovery, D.J. Abraham (ed.). Wiley, Hoboken, NJ, pp. 1–47,
2003.
23. Doweyko, A.M., QSAR: Dead or alive? J. Comput.-Aided Mol. Des., 2008, 22(2): 81–89.
24. Poling, B., J.M. Prausnitz, and J.P. O’Connell, Properties of Gases and Liquids (5th
edition). McGraw-Hill Book Company, United States, 2001.
25. Devillers,J. andA.T. Balaban, (eds). Topological Indices and Related Descriptors in QSAR
and QSPR. Academic Press, Singapore, p. 811, 2000.
26. Fredenslund, A., J. Gmehling, and P. Rasmussen, Vapor–Liquid Equilibrium Using
UNIFAC. A Group-Contribution Method. Elsevier: Amsterdam, the Netherlands, 1977.
27. Wang, S. and G.W.A. Milne, Graph theory and group contributions in the estimation of
boiling points. J. Chem. Info. Comput. Sci., 1994, 34: 1242–1250.
28. Faulon, J.-L., J. Visco, D.P., and D. Roe, Enumerating molecules. In: Reviews in Com-
putational Chemistry, K.B. Lipkowitz, R. Larter, and T.R. Cundari (eds). Wiley-VCH:
Hoboken, NJ, pp. 209–286, 2005.
29. Gordeeva, E.V., M.S. Molchanova, and N.S. Zefirov, General methodology and computer
program for the exhaustive restoring of chemical structures by molecular connectivity
indexes. Solution of the inverse problem in QSAR/QSPR. Tetrahedron Comput. Methodol.,
1990, 3: 389–415.
30. Skvortsova, M.I., et al., Inverse problem in QSAR/QSPR studies for the case of topolog-
ical indices characterizing molecular shape (Kier indices). J. Chem. Info. Comput. Sci.,
1993, 33: 630–634.
31. Kier, L.B., L.H. Hall, and J.W. Frazer, Design of molecules from quantitative structure–
activityrelationship models. I. Information transfer between path and vertex degree counts.
J. Chem. Info. Comput. Sci., 1993, 33: 143–147.
32. Kier, L.B., L.H. Hall, and J.W. Frazer, Design of molecules from quantitative structure–
activity relationship models. II. Derivation and proof of information transfer relating
equations. J. Chem. Info. Comput. Sci., 1993, 33
: 148–152.
33.
Skv
ortsova, M.I., et al., Molecular design of chemical compounds with prescribed prop-
erties from QSAR models containing the Hosoya index. Internet electron. J. Mol. Des.,
2003, 2: 70–85.
34. Skvortsova, M.I., et al., Inverse problem in the structure–property relationship analysis
for the case of information toplogical indices. Doklady Chem., 1997, 357: 252–254.
292 Handbook of Chemoinformatics Algorithms
35. Constantinou, L. and R. Gani, New group contribution method for estimating properties
of pure compounds. AIChE J., 1994, 40(10): 1697–1710.
36. Gani, R., B. Nielsen, andA. Fredenslund,Agroup contributionapproach tocomputer-aided
molecular design. AIChE J., 1991, 37(9): 1318–1332.
37. Chen, B., et al.,Application of CAMD in separating hydrocarbons by extractivedistillation.
AIChE J., 2005, 51(12): 3114–3121.
38. Patkar, P.R. and V. Venkatasubramanian, Genetic algorithms based CAMD. In: Com-
puter Aided Molecular Design: Theory and Practice, L.E.K. Achenie, R. Gani, and V.
Venkatasubramanian (eds). Elsevier: Amsterdam, the Netherlands, pp. 95–128, 2003.
39. Venkatasubramanian, V., K. Chan, and J.M. Caruthers, Computer-aided molecular design
using genetic algorithms. Comput. Chem. Eng., 1994, 18(9): 833–844.
40. Harper, P.M., M. Hostrup, and R. Gani, A hybrid CAMD method. In: Computer Aided
Molecular Design, L.E.K. Achenie, R. Gani, and V. Venkatasubramanian (eds). Elsevier:
Amsterdam, pp. 129–165, 2003.
41. Gillett, V.J., et al., SPROUT, HIPPO, and CAESA: Tools for de novo structure gener-
ation and estimation of synthetic accessibility. Perspect. Drug Discov. Des., 1995, 3:
pp. 34–50.
42. Harper, P.M. and R. Gani, A multi-step and multi-level approach for computer aided
molecular design. Comput. Chem. Eng., 2000, 24: 677–683.
43. Harper, P.M., et al., Computer-aided molecular design with combined molecular modeling
and group contribution. Fluid Phase Equi., 1999, 158–160: 337–347.
44. Hall, L.H. and L.B. Kier, Molecular connectivity chi indices for database analysis and
structure–property modeling. In: Topological Indices and Related Descriptors in QSAR
and QSPR, J. Devillers and A.T. Balaban (eds). Gordon and Breach Science Publishers:
Canada, pp. 307–360, 1999.
45. Kier, L.B., et al., Molecular connectivity I: Relation to non-specific local anesthesia. J.
Pharm. Sci., 1975, 64: 1971–1974.
46. Kier, L.B., Indexes of molecular shape from chemical graphs. Acta Pharm. Jugosl., 1986,
36: 171.
47. Kier, L.B., A shape index from molecular graphs. Quant. Struct. Act. Relat., 1985, 4: 109.
48. Kier, L.B. and L.H. Hall, Molecular Structure Description. Academic Press: San Diego,
CA, 1999.
49. Gani, R., C. Jimenez-Gonzalez, and D.J.C. Constable, Method for selection of solvents
for promotion of organic reactions. Comput. Chem. Eng., 2005, 29: 1661–1676.
50. Churi, N. and L.E.K. Achenie, Novel mathematical programming model for computer
aided molecular design. Ind. Eng. Chem. Res., 1996, 35: 3788–3794.
51. Odele, O. and S. Macchietto, Computer aided molecular design: A novel method for
optimal solvent selection. Fluid Phase Equi., 1993, 82: 47–54.
52. Ostrovsky, G.M., L.E.K. Achenie, and M. Sinha,
A reduced dimension branch-and-bound
algorithm
for
molecular design. Comput. Chem. Eng., 2003, 27: 551–567.
53. Sahinidis, N.V., M. Tawarmalani, and M.Yu, Design of alternative refrigerants via global
optimization. AIChE J., 2003, 49(7): 1761–1775.
54. Hall, L.H. and L.B. Kier, The molecular and connectivity chi indexes and kappa shape
indexes in structure–property modeling. In: Reviews in Computational Chemistry, K.B.
Lipkowitz and D.B. Boyd (eds). VCH Publishers: New York, pp. 367–422, 1991.
55. Siddhaye, S., et al., Pharmaceutical product design using combinatorial optimization.
Comput. Chem. Eng., 2004, 28: 425–434.
56. Ahuja, R.K., T.L. Magnanti, and J.B. Orlin, Network flows. In: Handbooks in Operations
Research and Management Science. G. L. Nemhauser, A. H. G. Rinnooy Kan, and M. J.
Todd, (eds). Elsevier: Amsterdam, the Netherlands, pp. 211–360, 1989.
Computer-Aided Molecular Design 293
57. Faulon, J.-L., On using graph-equivalent classes for the structure elucidation of large
molecules. J. Chem. Info. Comput. Sci., 1992, 32: 338–348.
58. Faulon, J.-L., Stochastic generator of chemical structure. 1. Application to the structure
elucidation of large molecules. J. Chem. Info. Comput. Sci., 1994, 34: 1204–1218.
59. Faulon, J.-L., D.P. Visco Jr., and R.S. Pophale, The signature molecular descriptor. 1.
Extended valence sequences and topological indices. J. Chem. Info. Comput. Sci., 2003,
43: 707–720.
60. Visco, J.D.P., et al., Developing a methodology for an inverse quantitative structure–
activity relationship using the signature molecular descriptor. J. Mol. Graph. Model.,
2002, 20: 429–438.
61. Weis, D., et al., The signature molecular descriptor. 5. The design of hydrofluoroether
foam blowing agents using inverse-QSAR. Ind. Eng. Chem. Res., 2005, 44: 8883–8891.
62. Faulon, J.L., http://www.cs.sandia.gov/∼jfaulon/QSAR/downloads.html
63. Weis, D., Inverse QSAR for compound development using the signature descriptor:
Application to hydrofluoroethers and γ-secretase inhibitors, in Department of Chemical
Engineering. Tennessee Technological University: Cookeville, TN, 2005.
64. Contejean, E. and H. Devie, An efficient incremental algorithm for solving systems of
linear diophantine equations. J. Chem. Inf. Comput. Sci., 1994, 113: 143–172.
65. Jackson, J.D., D.C. Weis, and D.P. Visco, Potential glucocorticoid receptor ligands with
pulmonary selectivity using I-QSAR with the signature molecular descriptor. Chem. Biol.
Drug Des., 2008, 72: 540–550.
66. Lipinski, C.A., et al., Experimental and computational approaches to estimate solubility
and permeability in drug discovery and development settings. Adv. Drug Deliv. Rev., 1997,
23: 3–25.
67. Faulon, J.-L., W.M. Brown, and S. Martin, Reverse engineering chemical structures from
molecular descriptors: How many solutions? J. Comput-Aided Mol. Des., 2005, 19(9–10):
637–650.
68. Faulon, J.-L., On using graph-equivalent classes for the structure elucidation of large
molecules. J. Chem. Info. Comput. Sci., 1992, 32: 338–348.
10
Computer-Aided
Molecular Design
De Novo Design
Diana C. Roe
CONTENTS
10.1 Introduction.....................................................................295
10.2 Historical Approaches to De Novo Design ....................................297
10.2.1 Identify Interaction Sites ..............................................298
10.2.2 Molecule Building Blocks .............................................298
10.2.3 Structure Generation and Search Strategies ..........................299
10.2.4 Structure Evaluation ...................................................300
10.2.5 Synthetic Accessibility and ADMET .................................301
10.3 Common Algorithms in De Novo Structure Generation......................302
10.3.1 Grow....................................................................302
10.3.1.1 Programs in Current Use that Implement Grow ...........304
10.3.1.2 Advantages, Limitations, and Computational
Complexity? .................................................306
10.3.2 Fragment-Link .........................................................306
10.3.2.1 Programs in Current Use that Implement
Fragment-Link ...............................................307
10.3.2.2 Advantages, Limitations, and Computational
Complexity? .................................................307
10.3.3 Sampling Strategies with EAs.........................................308
10.3.3.1 Programs in Current Use that Implement Sampling
Strategies.....................................................309
10.3.3.2 Advantages, Limitations, and Computational
Complexity? .................................................309
10.4 Summary .......................................................................310
References ............................................................................310
10.1 INTRODUCTION
A vital goal in drug discovery is identifying novel compounds that can serve as a
starting point in drug discovery. It is estimated that there are 10
60
−10
100
[1,2] potential
chemical compounds that each have a molecular weight of less than 500 Da. By
295
296 Handbook of Chemoinformatics Algorithms
comparison, PubChem, the largest database of known chemicals, has a little more than
19 million compounds to date [3], covering only a very small percentage of potential
chemical space. Even combinatorial libraries, which can range in size up to billions
of compounds, do not begin to fully sample the range of all possible compounds.
As the full compound space is too vast to search comprehensively, strategies have
to be employed to search this space efficiently for the discovery of novel lead drug
compounds.
De novo design handles this challenge by building compounds from scratch to
complement the target receptor. The guiding principle in this approach is that small
molecules that are complementary to the target receptor, both in shape and chemical
properties, will have the most specific binding. Resulting compounds also need to
be “drug-like” and readily synthetically accessible. In theory, any molecule of chem-
ical space could be constructed using de novo approaches. To reduce the search of
chemical space to a manageable problem, strong physical constraints must be taken
into account at each step during the generation of the lead drug molecule, limiting
the chemical space explored to those regions specifically complementary to the target
receptor. The advantage of this approach over a virtual screening strategy to identify
these compounds is that the search is directed to the relevant regions in chemical
space with a far greater range and diversity of potential lead compounds that can be
evaluated. Also, since compounds are built within the shape constraints of the tar-
get receptor, the structures are generated with optimal conformational geometries for
binding. In most virtual screening algorithms these conformations must be sampled
and can be missed. The main drawback is that the resulting compounds need to be
experimentally synthesized and tested, rather than taken from an in-house inventory
or ordered commercially. As our ability to predict binding affinities improves, the
trade-off between greater speed of screening and greater diversity of results may
drive an increase in use of de novo design strategies.
De novo design is inherently combinatoric, as many choices are available at each
step in molecule generation, leading to a non-polynomial (NP)-hard problem that
cannot be provably solved to the global optimum. Any solution to the problem is
going to represent a local optimum. And so the success of this approach depends
on well-chosen constraints for the problem and an efficient search strategy. The pri-
mary constraints are the geometric and chemical constraints derived from the target
receptor or target ligand(s), and internal constraints for the geometry and chem-
istry of the lead compound being constructed. The shape and chemical constraints
include both positive and negative requirements. For example, receptor site points or
“hot spots” of interaction must be matched with complementary functional groups in
all candidate structures, whereas boundary constraints that define disallowed struc-
tural regions must be avoided. These primary constraints are directly handled in
the structure generation phase in de novo design. The secondary constraints include
synthetic accessibility of the final compounds and their predicted adsorption, distri-
bution, metabolism, elimination, and toxicity (ADMET) properties. These constraints
are handled both by heuristics employed during structure generation and also as filters
on the final set of compounds. While the set of constraints are similar to all de novo
design algorithms, strategies for generating structures and for searching chemical
space to fit these constraints vary considerably among the programs.
Computer-Aided Molecular Design 297
One major distinction in de novo design programs is whether they are receptor
based or ligand based. In receptor-based programs the three-dimensional (3D) struc-
ture of the targetreceptor is knownand provides the primary constraints. Ligand-based
programs either generate a 3D structural pharmacophore model to generate geomet-
ric and chemical constraints that are similar to the receptor-based constraints or they
use similarity to a known active ligand or QSAR model as the primary constraint.
The features in de novo design algorithms can be divided into the following compo-
nents, and choices at each of these components can distinguish one program from
another:
i. Site points: Site points represent “hot spots” of interaction with the target
receptor. They define the primary geometric and chemical constraints for
the structure generation. Some ligand-based de novo design programs use
similarity to a molecule template instead of site points.
ii. Molecule building blocks: These are the units for constructing the structure.
They can be atoms, fragments, or templates (generic fragments).
iii. Structure generation: The strategy chosen here is one of the ways used to
classify de novo design programs. Strategies include (a) “grow,” which starts
from a seed point and grows a structure; (b) “fragment-link,” where fragments
are placed in site points and linked together; and (c) sampling approaches,
where molecules are randomly grown. Implicit in each strategy is rules and
geometric constraints to generate chemically reasonable structures.
iv. Search strategy: The strategy used to search through the combinatoric set
of possible (sub)structures, combined with the structure generation strategy,
forms the core of the de novo design algorithm. Common search strategies
include breadth-first search, depth-first search, evolutionary algorithms (EAs),
and Monte Carlo.
v. Structure evaluation: Evaluates structures based on primary constraints—
geometry and chemistry for binding, or similarity to a known active ligand. By
evaluating substructures at every step during structure generation, the choices
can be pruned during structure generation.
vi. ADMET and synthetic accessibility: These are the secondary constraints that
can be taken into consideration as a scoring function during structure genera-
tion, as a postfilterafter construction. In addition, some programs use heuristics
during structure generation to incorporate these constraints.
While there are many variations on the algorithms at each of these component steps,
as described below, this review will focus on the structure generation algorithms.
10.2 HISTORICAL APPROACHES TO DE NOVO DESIGN
The first de novo design programs were receptor based. The field began with programs
being developedto describe the binding properties of a target receptor. Initial programs
[4–8] devised strategies to identify sites that represented “hot spots” of interaction
within a receptor, placed small molecule fragments or skeletons to interact at these