Suppose I have a set of small molecules (multiple chemotypes or clusters) with experimental activities (e.g., IC50s) for
several different assays. I can also calculate physical properties for each of the molecules. Does anyone have experience
attempting to perform simultaneous clustering using both fingerprints (to group compound structures) and activities from
multiple assays (to group biological profiles)? Obviously the numbers and distance metrics are not the same. What are
some of the approaches that can and should be used to fuse together this different information (if possible) so that
the groupings of molecules into clusters simultaneously reflects both structure similarity (and differences) as well as
property or activity similarity (and differences) with appropriate weighting for each of these kinds of information?
Also, has anyone tried to "project" this kind of information into a low (2)D space e.g., using MDS or something similar?
Can you get this to work reasonably well? What kind of scaling or adjustment of the variables was necessary?
If anyone has any datasets (even fake data) with a protocol they can share, that would be excellent.
Thank you.
Regards,
Jim Metz
(847) 936-0441
