The Use of GADriven Evolutionary Experiments

As shown above, GAs are rather tolerant to experimental errors and may still yield good re suits even if the starting hypothesis is wrong. This behavior is especially true for false nega tive results, since these are simply eliminated in the selection step and are not remembered The elimination of misleading false positive results takes somewhat longer - depending or, how often a good chromosome is allowed to replicate. This "fuzzy" and robust optimization property makes GAs especially attractive for real time experimental optimizations as described above. Furthermore, for the application of GA-based evolutionary methods, similar to Nature, one does not need to know or to describe chemical structures explicitly, or to develop a structure-based hypothesis about SARs.

On the other hand, GAs also have serious drawbacks which could render them unattractive for real experiments. First, GAs are stochastic which means that they do not necessarily give the same results in each run. By its very nature, a GA "learns" in a sequential, implicit way over several cycles of selection and testing which could be prohibitive for time-consuming, or expensive synthesis, or biological testing. Therefore, the application of GA-driven compound selection seems appropriate especially if [21,22]:

• the search space is very large and multi-dimensional,

• the sequential synthesis and testing can be performed rapidly, and

• the structure-property relationship is multimodal.

Keeping these criteria in mind, we believe that there are many SPR problems that could be solved with success using GA-based compound selection. Such problems could involve more complicated selection criteria that include not only biological activities but also other properties like selectivity or penetration through biological membranes, which would increase the dimensionality of the SPR landscape. Other examples may include finding entirely new multi-component reactions that yield products exhibiting biological activities simply by varying various starting materials. The area of material sciences where new materials or new catalysts are composed by mixing predefined starting materials or components, and explicit descriptions of SPR are rare, seems amenable to evolutionary selection methods. A genetic algorithm has recently been used to propose new polymer molecules that mimic a given target polymer [23], or exhibit a range of glass transition temperatures and hydrophobic-ities [24], Although the number of polymers considered was not very large, this work represents a convincing example of diversity and similarity selection that was verified on the basis of experimental data in the field of materials discovery.

[1] W. Fontana, P. Schuster, Science 1998 280,1451-1455.

[2] I. Ugi, M. Wochner, E. Fontain, J. Bauer, B. Gruber, R. Karl, in M. A. Johnson,

G. M. Maggiora (Eds.) Concepts and Applications of Molecular Similarity, John Wiley & Sons Inc., New York 1990, pp. 239-288.

[3] W. J. Cook, W. H. Cunningham, W. R. Pulleyblank, A. Schrijver, Combinatorial Optimization, Wiley, London 1997.

[4] J. H. Holland, Adaptation in Natural and Artificial Systems, The University of Michigan Press, Ann Arbor, MI 1975.

[6] L. Weber, Drug Discovery Today 1998,3, 379-385.

[7] J. Heitkoetter, D. Beasley, The Hitch-Hiker's Guide to Evolutionary Computation: A list of Frequently Asked Questions (FAQ) 1999. USENET: Available via anonymous FTP from:

[8] T. Baeck, D. B. Fogel, Z. Michalewicz (Eds.) Handbook of Evolutionary Computation, IOP Publishing and Oxford University Press, Bristol/New York 1997.

[9] J. H. Holland, Hidden Order - How Adaptation Builds Complexity, Addison-Wesley, Reading, MA 1996.

[10] M. Forrest, M. Mitchell, in Relative building-block fitness and the building-block hypothesis, in Foundations of Genetic Algorithms 2, D. Whitley (Ed.), Morgan Kaufmann, San Mateo, CA 1993, pp. 109-126.

[11] C. Stephens, H. Waelbroeck, Evolutionary Computation 1999, 7,109-124.

[12] R. Wehrens, E. Pretsch, L. M. C. Buydens, J. Chem. Inf. Comput. Sei. 1998,38,151-157.

[13] Y. Yokobayashi, K. Ikebukuro, S. McNiven, I. Karube, J. Chem. Soc. Perkin Trans. I 1996, 2435-2437.

[14] J. Singh, M. A. Ator, E. P. Jaeger, M. P. Allen, D. A. Whipple, J. E. Soloweij, S. Chowdhary, A. M. Treasurywala, /. Am. Chem. Soc. 1996,118,1669-1676.

[15] L. Weber, S. Wallbaum, C. Broger, K. Gubemator, Angew. Chem. Int. Ed. Engl. 1995 107, 2453-2454.

[16] K. Illgen, T. Enderle, C. Broger, L. Weber, Chemistry & Biology, 2000, 7,433-441.

[17] L. Weber, M. Almstetter, Diversity in very large libraries, in Molecular Diversity, R. Lewis (ed.), Kluwer, Amsterdam 1999.

[18] L. Weber, C. Broger, Hoffmann-La Roche AG, unpublished results.

[19] L. Weber, K. Illgen, M. Almstetter, SYNLETT1999,3,366-374.

[20] I. Rechenberg, Evolutionsstrategie: Optimierung technischer Systeme nach Prinzipien der biologischen Evolution, Frommann-Holzboog, Stuttgart 1973. 2nd edition 1994.

[21] G. Schneider, P. Wrede, Mathematical Res. 1993,81, 335-346.

[22] G. Schneider, W. Schrödl, G. Wallukat, E. Nissen, G. Rönspeck, J. Müller, P. Wrede, R. Kunze, Proc. Natl. Acad. Sei. USA 1998, 95,12179-12184.

[23] V. Venkatasubramanian, K. Chan, J. Caruthers, J., J. Chem. Inf. Comput. Sei. 1995, 35,188-195.

[24] C. H. Reynolds, J. Comb. Chem. Sei. 1999,1,297-306.

Virtual Screening for Bloactlve Molecules

Edited by Hans-Joachim Böhm and Gisbert Schneider Copyright© VV1LEY-VCH Verlag GmbH, 2000

Was this article helpful?

0 0

Post a comment