clustering versus self-organizing maps

Is anyone aware of any cheminformatics publications that clearly demonstrate good agreement between

clustering and self-organizing maps (SOMs).  In other words, if a set of molecules is in cluster # 1, I would

expect them to exist in either the same SOM grid point, or a close set of grid points. I would not expect

the molecules in cluster # 1 to be spread out in a SOM.

If someone has a PP protocol that they are willing to share that demonstrates good agreement between

these two approaches that would be helpful.  So far, I am not able to demonstrate what I intuitively

expect to be reasonable agreeement using the same input set of molecules and the same set

of molecular descriptors for clustering and for SOM generation, but ... perhaps I am not doing

something correctly!

Details - I am processing a set of ~ 1500 molecules, descriptors are: ECFP_6, Num_Rings,

Num_AromaticRings, Num_RotatableBonds.  I am using the R stats SOM component with

a rectangular grid size of 30 x 30, and the PP cluster molecules component - again using

exactly the same set of molecular descriptors.

Thank you.

Regards,

Jim Metz

James.Metz@AbbVie.com