Binning scheme

Bayley, M. J. and Willett, P. (1999) Binning schemes for partition-based compound selection../. Mol. Graph. Model. 17, 10-18. 

The ChemDiverse method uses predefined ranges for measuring the distances between the points (pharmacophoric features) distances are calculated exactly but stored using this binning scheme, with each distance represented by the bin into whose range it falls. Each pharmacophore for the 4-point method needs six distances to be characterised (to form a tetrahedron), whereas three distances are needed for each 3-point pharmacophore (triangle). All the combinations of features and distances, combined for all the evaluated conformers, are stored in a pharmacophore key . 

It was demonstrated that to obtain the best surface autocorrelation vectors for QSAR modeling, the van der Waals surface is better than other molecular surfaces. Then, surface should have no fewer than five grid points per A, and a distance interval not more than 1A should be used in the distance binning scheme (Figure Al). 

There are different binning schemes they usually satisfy two simple criteria [Bayley and Willett, 1999] (1) the maximum and minimum values for each of the descriptors that specify the partition must be set so as to encompass all of the compounds that may need to be processed by the partitioning scheme and (2) it seems appropriate that each molecule be assigned to just a single cell, thus requiring that the cell ranges do not overlap at all. 

Figure Cl Example of the four binning schemes from [Bayley and Willett, 1999]. (a) Original data represented by Di — D2 descriptors (b) equisized independent binning scheme (c) equifrequent independent binning scheme (d) equisized dependent binning scheme (e) equifrequent dependent binning scheme.

Several cell-based diversity measures have been proposed in the literature [Pascual, Borrell et al, 2003 Pascual, Mateu et al, 2003]. These are concentration indices, such as statistics, Gini concentration ratio, and Pratt measure, used to evaluate the distribution of compounds throughout the grid specified by a binning scheme and, thus, sometimes referred to as occupancy numbers. 

Table II lists the proteins used in the training set, a subset of the PDB Select database of nonhomologous proteins [21]. We avoided inclusion of proteins that form dimers (or other oligomers) in solution because one would expect the distributions in this case to be significantly altered due to the oligomerization process. For each protein we list the PDB code, number of residues, radius of gyration, and classification in our size bin scheme.

---