PKOM: A tool for clustering, analysis and comparison of big chemical collections

We describe the algorithm underlying PKOM, a methodology for clustering, analysis and visualization of multi-dimensional data onto a two-dimensional map. PKOM is based on the mixture of two very popular methods that have been widely used by the pharmaceutical industry for the clustering of genomic or SAR (Structure Activity Relationship) chemical information. The first method at the origin of PKOM is SOM (Self-Organizing Maps), a clustering technique based on neural networks. The second method is TREE MAPS, a visualization method based on hierarchical clustering by dendrograms. We initially describe herein the two methods and the reasons why we have taken the best of both to merge them into PKOM. We then describe in detail the PKOM algorithm and its advantages compared to the two former. Examples are given on how to apply this kind of 2-D topological clustering technique to the organization of big pharmaceutical collections in practical cases.