Hello,
I've refined a structure with CNS to Rwork/free=0.226/0.273. I switched to
Refmac5.2 to take advantage of TLS refinement and set up a run with 10
cycles of TLS refinement (groups as suggested by the TLSMD server) followed
by 10 cycles of restrained co-ordinate refinement. After the TLS cycles
the model was improved (Rwork/free=0.222/0.241), however during co-ordinate
refinement Rfree refined up (final Rwork/free=0.197/0.264). My
understanding would be that the TLS refinement is modelling the ADPs most
accurately, but that Refmac's co-ordinate refinement is over-fitting
slightly. Would this seem correct? And if so, is it valid to run Refmac
with no cycles of co-ordinate refinement and take the resulting model as
the final structure?
Thanks very much,
Nick
----------------------
NM Burton, Biochemistry
[log in to unmask]
|