Improving Similarity-Driven Library Design: Customized Matching and Regioselective Feature Trees

Abstract

Reduced graph descriptors, like feature trees, are frequently applied in cases where the relative arrangement of functional groups is more important than exact substructure matches. Due to their ability to deal with fragmented molecules, they are well-suited for fragment space search and library design. We recently presented LoFT, a novel focused library design approach based on feature trees. During evaluation two drawbacks of the reduced graph descriptor were discovered: First, regioisomeric substructures cannot be distinguished in feature tree mappings which results in a large information loss especially when connecting R-groups to cores. Second, the automatic matching procedure might result in undesired alignments, since the knowledge on what is considered as core by the user is not taken into account. In the following, we will present two approaches to overcome those drawbacks. The generation of the feature trees is modified, so that different arene substitution patterns can be recognized and a customized matching is introduced, allowing the user to determine the parts of the query, where the reagents are allowed to match. Subsequently we investigate the improvements on library design by reviewing the design scenarios which were already used for the evaluation of LoFT.