We introduced Refined Nearest Neighbor analysis methods for the analysis of chemcial datasets in the paper:

Rohrer, S.G.; Baumann, K.
Impact of Benchmark Data Set Topology on the Validation of Virtual Screening Methods: Exploration and Quantification by Spatial Statistics.
J. Chem. Inf. Model., 2008, 48, 704-71
doi: 10.1021/ci700099u (Open Access)

It was shown, that datasets with a "clumpy" topology in in descriptor space bias validations of virtual screening methods towards over-optimistic results. This lead to the rationale, that datasets without bias, i.e. Maximum Unbiased Validation (MUV) Datasets, should have a non-clumpy, spatially random topology.

Both the original findings about the impact of dataset topology and the rationale behind MUV dataset design are summarized quite concisely in a talk Knut Baumann gave at the EuroQSAR 2008 at Uppsala: Slides.

Please refer to muv@tu-bs.de if you have any questions or suggestions.