I have generated a set of fragments from a set of ~2000 compounds using the RingAssemblies option in the "Generate Fragments" component with the intention of using these in a substructure search against a large library.  I've included alpha atoms to carry through chirality information.  If I take the raw set of generated fragments into "Substructure Filter From Tag", I get ~12K hits in the large library.  If I put the generated fragments through "Merge Molecules" first (with the intention of generating a unique set of fragments and speeding up the search) I only get ~9K hits. 

If I test the merged fragments and use them in a substructure search against the 2000 set, the merged fragment set sometimes fails to find compounds which generated that fragment.  Merge Molecules isn't doing quite what I want, and given that these aren't real molecules, I can see it might not be designed for this .  (I've set it to "append or create" and "append if new".)  Is there a way to get the unique fragments?

thanks