Selected Examples of New Approaches to Molecular Similarity

Despite the wide range of existing methods for describing molecular similarity, new techniques are under continuous development. Their major goal is an improved description of molecular similarity, while maintaining computation speed. 

As a quantitative performance measure of a similarity descriptor, the number of molecules from a database belonging to the same class as a reference molecule can be used. The enrichment factor is such a measure, and this takes the ratio of hits to the entire database size into account. It is defined as number of hits in the first percent of the database sorted according to the similarity measure, divided by the expected number of hits retrieved with a random selection. Intuitively, the enrichment factor describes how much is gained by using a similarity measure for selecting molecules compared to a random selection. For the design of diverse libraries, the detection of molecular classes is a necessary feature of the similarity measure. 

Besides the quantitative measure of the number of retrieved hits, their quality in the sense of usefulness for lead discovery is also of interest. Molecules bearing a common scaffold are typically less useful, as they do not reveal a chemically new motif. Therefore, 

The main difficulty for commonly used 3D descriptors results from the treatment of conformational flexibility. Considering that only a single conformation is inadequate for most classes of pharmaceutically relevant molecules, averaging over conformational space provides only a very rough view as to what is intended to be described, namely the possible arrangement of pharmacophoric groups in Cartesian space. In this section a more detailed view on three subjectively selected approaches to describe molecular similarity will be presented. The approaches have in common the fact that similarity is quantified to a less dependable degree from only the 2D structure or specific 3D conformations. 

---