Use of Markush structure analysis techniques for descriptor generation and clustering of large combinatorial...

This article outlines a system in which a Markush structure representation is used for analysis of a combinatorial library, giving savings in terms of storage and processing requirements when compared with an enumerated library. The Markush representation used is described, along with the way in which it can be built from a reaction- and precursor-based description of a library. The process used to generate SMILES, and structure fingerprint and calculated property descriptors for each specific molecule in the library is discussed. Comparative performance figures are given for the Markush approach and various standard fingerprint generation programs based on the enumerated members of the library; the former shows time savings of a couple of orders of magnitude. The use of these rapidly generated fingerprints for clustering of large combinatorial libraries is described.