Neighborhood Behavior of in Silico Structural Spaces with Respect to In Vitro Activity Spaces-A Benchmark for Neighborhood Behavior Assessment of Different in Silico Similarity Metrics

In a previous work, we have introduced Neighborhood Behavior (NB) criteria for calculated molecular similarity metrics, based on the analysis of in vitro activity spaces that simultaneously monitor the responses of a compound with respect to an entire panel of biologically relevant receptors and enzymes. Now, these novel NB criteria will be used as a benchmark for the comparison of different in silico molecular similarity metrics, addressing the following topics: (1) the relative performance of 2D vs 3D descriptors, (2) the importance of the similarity scoring function for a given descriptor set, and (3) binary or Fuzzy Pharmacophore Fingerprints-can they capture the similarity of the spatial distribution of pharmacophoric groups despite different molecular connectivity? It was found that fuzzy pharmacophore descriptors (FBPA) displayed an optimal NB and, unlike their binary counterparts, were successful in evidencing pharmacophore pattern similarity independently of topological similarity. Topological FBPA, identical to the former except for the use of topological instead of 3D atom pair distances, display a somehow weaker, but significant NB. Metrics based on "classical" global 2D and 3D molecular descriptors and a Dice scoring function also performed well. The choice of the similarity scoring function is therefore as important as the choice of the appropriate molecular descriptors.