Metric Validation and the Receptor-Relevant Subspace Concept

Following brief comments regarding the advantages of cell-based diversity algorithms and the selection of low-dimensional chemistry-space metrics needed to implement such algorithms, the notion of metric validation is discussed. Activity-seeded, structure-based clustering is presented as an ideal approach for the validation of either high- or low-dimensional chemistry-space metrics when validation by computer-graphic visualization is not possible. Whereas typical methods for reducing the dimensionality of chemistry-space inevitably discard potentially important information, we present a simple yet novel algorithm for reducing dimensionality by identifying which axes (metrics) convey information related to affinity for a given receptor and which axes can be safely discarded as being irrelevant to the given receptor. This algorithm often reveals a three- or two-dimensional subspace of a (typically six-dimensional) BCUT chemistry-space and, thus, enables computer graphic visualization of the actual coordinates of active compounds and combinatorial libraries. Most significantly, we illustrate the importance of using receptor-relevant distances for identifying near neighbors of lead compounds, comparing libraries, and other diversity-related tasks.