On 3/1/11 12:23 AM, Ashish Runthala wrote: > NEVER USE DOPE ALONE TO RANK MODELS. ONLY DOPE IS NOT A GOOD SELECTION MEASURE.
Such a strong statement needs to be backed up by a publication... particularly given that it is contradicted by both your next sentence and our own study (Prot Sci 17, 1881-1893, 2008).
> Use normalized_dope
Normalized DOPE is derived from the raw DOPE score. Thus the ranking of models, given the same sequence, will be the same with DOPE or normalized DOPE.
However, since raw DOPE is an "energy", you should certainly not use it to rank models that differ in sequence. For that, normalized DOPE would be more appropriate.
> or GDT to rank models/Clusters/Hybrid Models.
GDT is a structural measure, like RMSD, so it requires another structure to compare against. This makes it useless for assessing comparative models unless you already know the answer (as in CASP).
Ben Webb, Modeller Caretaker