Combinatorial libraries with virtual screening

Closely related to the use of PSA in virtual screening is its application in the design of combinatorial libraries with optimal properties. These applications are reviewed further in Refs. [46, 47], for example. 

Lobanov, V. S. (2000) High-throughput screening of virtual combinatorial libraries with neural networks. Abstr. Pap. Am. Chem. Soc. 220th, CINF-062. 

Software tools for virtual screening can be best classified by the input data available for screening. On the one side, there is always a collection of compounds to be screened, which differs in size (from a few tens to several millions) and in structure (from structurally unrelated compounds via combinatorial libraries to chemistry spaces). On the other side, there is the data that is used to create the screening query, which can be a protein structure, a known active compound or a pharmacophore created from several known actives (see Figure 4.1). In summary, we are ending up with four classes of screening tools ... 

Fig. 4.1 Virtual screening tools can be categorized by the compound data to be screened (compound collection, combinatorial library, chemistry space) and the query type (structure-based, ligand-based, descriptor-based, pharmacophore-based). The output is always a list of compounds together with a score quantifying the fit to the query.

In general, reagent-based selection is much faster and more convenient to execute in the laboratory as compared with the product-based selection. On the other hand, the latter strategy usually provides more accurate results. There exists a potential to combine both approaches to achieve more optimal results, particularly in the case of large exploratory virtual combinatorial libraries, for which mass random synthesis and screening are not economically feasible. In this article, we demonstrated the usefulness of property-based approach for selection of optimal GPCR ligands. 

Virtual screening on a 2-D basis is far more efficient in terms of computing time, although the information content of the resulting virtual hits is far less sophisticated than for a 3-D pharmacophore search. It is in this setting that combinatorial chemistry approaches are applied most effectively. The many interactions possible with a compound library based on virtual hits compensates for the inherent fuzziness of any prediction tool. Examples have been described where the 2-D structure of a... 

CLEVER is a computational tool designed to support the creation, manipulation, enumeration, and visualization of combinatorial libraries. The system also provides a summary of the diversity, coverage, and distribution of selected compound collections. When deployed in conjunction with large-scale virtual screening campaigns, CLEVER can offer insights into what chemical compounds to synthesize, and, more importantly, what not to synthesize. In this chapter, we describe how CLEVER is used and offer advice in interpreting the results. 

In collaboration with University of Trieste, we have developed rational approaches for the design and synthesis of peptidomimetic and non-peptidic inhibitors of HIV PR, utilizing structure-based [12-15], as well as combinatorial, library design methods [16, 17]. In this paper, we survey computer-assisted studies on the design, focusing and in silico screening of virtual combinatorial libraries of peptidomimetics and cyclic ureas, as potential anti-HIV agents, that were carried out in our laboratory. 

The multiple potential pharmacophore key calculated from a ligand can be compared to the multiple potential pharmacophore key of complementary site-points in its target binding site. This provides a novel method to measure similarity when comparing ligands to their receptors, with applications such as virtual screening and structure-based combinatorial library design. 

---