Issues in Molecular Diversity and the Role of Ligand Binding Sites.- Molecular Diversity in Drug Design. Application to High-speed Synthesis and High-Throughput Screening.- Background Theory of Molecular Diversity.- Absolute vs Relative Similarity and Diversity.- Diversity in Very Large Libraries.- Subset-Selection Methods For Chemical Databases.- Molecular Diversity in Site-focused Libraries.- Managing Combinatorial Chemistry Information.- Design of Small Libraries for Lead Exploration.- The Design of Small- and Medium-sized Focused Combinatorial Libraries.
Chapter 5
Diversity in Very Large Libraries (p. 93-94)
Diversity in Very Large Libraries
Lutz Weber and Michael Almstetter
Morphochem AG, Am Klopferspitz 19, 82152 Martinsried, Germany
Key words: Combinatorial chemistry, genetic algorithms, combinatorial optimisation, QSAR, evolutionary chemistry, very large compound libraries
Abstract: Combinatorial chemistry methods can be used, in principle, for the synthesis of very large compound libraries. However, these very large libraries are so large that the enumeration of all individual members of a library may not be practicable. We discuss here how one may increase the chances of finding compounds with desired properties from very large libraries by using combinatorial optimisation methods. Neuronal networks, evolutionary programming and especially genetic algorithms are heuristic optimisation methods that can be used implicitly to discover the relation between the structure of molecules and their properties. Genetic algorithms are derived from principles that are used by nature to find optimal solutions. Genetic algorithms have now been adapted and applied with success to problems in combinatorial chemistry. The optimisation behaviour of genetic algorithms was investigated using a library of molecules with known biological activities. From these studies, one can derive methods to estimate the diversity and structure property relationships without the need to enumerate and calculate the properties of the whole search space of these very large libraries.
1. INTRODUCTION
In nature, the evolution of molecules with desired properties may be regarded as a combinatorial optimisation strategy to find solutions in a search space of unlimited size and diversity. Thus, the number of all possible, different proteins comprising only 200 amino acids is 20200, a number that is much larger than the number of particles in the universe (estimated to be in the range of 1088 . ) Similarly, the number of different molecules that could be synthesised by combinatorial chemistry methods far exceeds our synthetic and even computational capabilities in reality. Whilst diversity and various properties of compound libraries in the range of several thousands to millions can be calculated by using a range of different methods, there is little available knowledge and experience for dealing with very large libraries. The task for chemists is therefore to find methods that can be used to choose useful subsets from this practically unlimited space of possible solutions.
The intellectual concept and the emerging synthetic power of combinatorial chemistry are moving the attention of experimental chemists towards a more abstract understanding of their science: instead of synthesising and investigating just a few molecules they are dealing now with libraries and group properties. The answers to questions such as how diverse or similar are any two compounds, are now not just intellectually interesting but also have commercial value. Therefore, the ability to understand and use very large libraries is, in our opinion, connected to the understanding and the development of chemistry in the future.
The discovery of a new medicine may be understood as an evolutionary process that starts with an initial knowledge set, elaborating a hypothesis, making experiments and thereby expanding our knowledge. A new refined hypothesis will give rise to further cycles of knowledge and experiments ending with molecules that satisfy our criteria. If very large compound libraries are considered, one may argue that the desired molecules are already contained within this initial library. A very large library on the other hand means that we are neither practically nor theoretically able to synthesise or compute all members of this library. How can we nevertheless find this molecule? Is it possible to develop methods that automate the discovery of new medicines by using such libraries without human interference?
An answer to these questions would be a novel approach to combinatorial chemistry that tries to connect the selection and synthesis of biologically active compounds from the very large library by mathematical optimisation methods. Heuristic algorithms, like genetic algorithms or neural networks, mimic the Darwinian evolution and do not require the a priori knowledge of structure-activity relationships. These combinatorial optimisation methods (1) have proved to be useful in solving multidimensional problems and are now being used with success in various areas of combinatorial chemistry. Thus, evolutionary chemistry may aid in the selection of information rich subsets of available compound libraries or in designing screening libraries and new compounds to be synthesised, adding thereby a new quality to combinatorial chemistry.
