USING A GENETIC ALGORITHM TO SUGGEST COMBINATORIAL LIBRARIES

In combinatorial synthesis, molecules are assembled by linking chemically similar fragments. Since the number of available chemical fragments often greatly exceeds the number of distinct fragments that can be used in one synthetic experiment, choosing a subset of fragments becomes problematical. For example, only a few dozen distinct primary and secondary amines have ever been reported to have been used in constructing a library of peptoids (oligomers of N-substituted glycine), while there are several thousand suitable primary and secondary amines that are commercially available. If a combinatorial library is to be constructed with a particular biological activity in mind, computer-based structure-activity methods can be used to rationally select a subset of fragments. In principle one would computationally generate every possible molecule as a combination of fragments, score each molecule by the Likelihood of its being active, and select those fragments that occur in high-scoring molecules. For many cases there are too many combinations to take this exhaustive approach, but genetic algorithms can be used to quickly find high-scoring molecules by sampling a small subset of the total combinatorial space. In this paper we demonstrate how a genetic algorithm is used to select a subset of amines for the construction of a tripeptoid library. We show three examples. In the first example, the scoring is based on the similarity of the tripeptoids to a specific tripeptoid target. Since the target itself can be generated in this example, we have an opportunity to experiment with the protocol of our genetic algorithm. In the second example, scoring is based on the similarity to two tetrapeptide CCK antagonists. In the third, scoring is done by a trend vector derived from activity data on ACE inhibitors. In all cases we show that the genetic algorithm can find, in a modest amount of computer time, high-scoring peptoids that resemble the targets.