ErG:  2D Pharmacophore Descriptions for Scaffold Hopping

Abstract

An extended reduced graph approach (ErG) is presented that uses pharmacophore-type node descriptions to encode the relevant molecular properties. The basic idea of the method can be described as a hybrid approach of reduced graphs (Gillet et al. J. Chem. Inf. Comput. Sci. 2003, 43, 338-345) and binding property pairs (Kearsley et al. J. Chem. Inf. Comput. Sci. 1996, 36, 118-127). However, specific extension modifications to correctly describe the pharmacophoric properties, size, and shape of the molecules under study result in a very stable and good performance as compared to DAYLIGHT fingerprints (DFP). This is exemplified for 11 activity classes of the MDL Drug Data Report database, for which ErG performs as well or better than DFP in 10 cases. On the basis of the example data sets, the ability of ErG to switch from one chemotype to another (often referred to as "scaffold hopping") is highlighted. Additionally, possible pitfalls of reduced graph approaches as well as suitable solutions are discussed with the help of example structures. Overall, it is shown that ErG is a widely applicable method capable of identifying structurally diverse actives for a given active search query. This diversity is achieved by a high degree of molecular abstraction, which in turn results in a low dimensional descriptor vector that allows very low computation times for similarity searches.

Chemoinformatics at the University of Sheffield 2002–2014

“…The substituents can be specific substructures, such as methyl or ethyl, or they can be defined using generic nomenclature, Since 2002, the Sheffield group has investigated the use of reduced graphs in a number of applications including similarity searching, analysis of HTS data, clustering and identification of structure-activity relationships. Our approach to representing structures as reduced graphs is close in concept to the Feature Trees developed by Rarey et al [47] and the ErGs presented by Steifl et al [48] Gillet et al…”

Indirect Similarity Based Methods for Effective Scaffold-Hopping in Chemical Compounds

“…Using the Best-Sum retrieval strategy (described in section 5.1.1) for the query q, the first compound retrieved is again c 6 . This compound is added to the set A and removed from the database.…”

“…The scaffold-hopping performance of a method is then evaluated by identifying how many active compounds from different clusters (structural classes) are obtained. 6 The assumption in this approach is that compounds with different scaffolds are more likely to be in different clusters and the method that retrieves actives from many different clusters among the top hits is good for scaffold-hopping. This evaluation strategy has the drawback that using global clustering masks the local effects and the final clusters depend on clustering parameters.…”

Reduced Graphs and Their Applications in Chemoinformatics

“…We recognise the substantial efforts made by other groups in related methods, most notably the feature trees approach by Rarey et al (1,2) and the extended reduced graph, ErG, by Stiefl et al (3,4), however, these are not discussed in detail for reasons of brevity. …”