Multiple-Ligand-Based Virtual Screening:  Methods and Applications of the MTree Approach

Abstract

We present a novel approach for ligand-based virtual screening by combining query molecules into a multiple feature tree model called MTree. All molecules are described by the established feature tree descriptor, which is derived from a topological molecular graph. A new pairwise alignment algorithm leads to a consistent topological molecular alignment based on chemically reasonable matching of corresponding functional groups. These multiple feature tree models find application in ligand-based virtual screening to identify new lead structures for chemical optimization. Retrospective virtual screening with MTree models generated for angiotensin-converting enzyme and the alpha1a receptor on a large candidate database yielded enrichment factors up to 71 for the first 1% of the screened database. MTree models outperformed database searches using single feature trees in terms of hit rates and quality and additionally identified alternative molecular scaffolds not included in any of the query molecules. Furthermore, relevant molecular features, which are known to be important for affinity to the target, are identified by this new methodology.

Computational Methods in Drug Discovery

“…As mentioned, one drawback is that all features of a query molecule are equally important for ranking candidate molecules, regardless of any effect of these features on the biologic activity at a target. Hessler et al (2005) proposed a method that combines the advantages of similarity and pharmacophore searching on the basis of 2D structural representations only. In their proposed method, a set of query molecules is converted into a topological model (MTree) based on chemically reasonable matching of corresponding functional groups.…”

“…In their proposed method, a set of query molecules is converted into a topological model (MTree) based on chemically reasonable matching of corresponding functional groups. This creates a topological map of the most similar fragments from a set of structurally diverse but 368 active molecules, and conserved features are characterized by high similarity scores of the corresponding nodes in the MTree model (Hessler et al, 2005). Because of the low dependence on chemical substructures, they argue that the MTree model is especially useful for identification of alternative novel molecular scaffolds or chemotypes.…”

Comparison of Structure‐ and Ligand‐Based Virtual Screening Protocols Considering Hit List Complementarity and Enrichment Factors

“…To compare reference and screening compounds, we applied the splitsearch, match-search and dynamic match-search algorithms. [5,36] In the virtual screening, all candidate ligands were ranked according to their resulting similarity to the respective reference ligand.…”

“…[3,[7][8][9] Different methodologies can make use of either 2D or 3D descriptors. Given 3D conformations of one or more active ligands derived from structure determination methods or from molecular modelling, 3D-similarity [5] or pharmacophore searches [10] represent an option for the virtual screening of compound libraries.…”

“…[32] The obtained docking poses were rescored and ranked with different scoring functions (see the "Experimental" Section). For ligand-based virtual screening, we applied 3D-similarity searches (using ROCS [7] ), and 2D-similarity searches using Feature Trees [5] or Scitegic Functional Fingerprints (FCFP4).…”