What's the best way of dealing with unknown atoms (UNK) in phenix.refine?

I have a long disordered loop that runs basically the length of the protein on the back side and there is a single "amino acid" in the middle that I'm trying to model as UNK.  I can run a round of refmac and auto generate a lib file with my ligand and the unk atoms but phenix.refine balks at that with respect to the Van der Waals radii.

Thanks for the help,

-Jeff