I read that the final step in the low-resolution part of a rosetta classic ab-initio protein fold prediction protol should be rescoring the obteined structures with the score4 in order to discriminate the native-like structures.
Looking for this score4 I could find the following score4-like scores (both in the pyrosetta package and rosetta3.4 package): “score4L”, score4_smooth”, score4_smooth_cart”.
In all papers I’ve read I could not find any references to any of them but only to “score4”. Does anybody know which this “score4” is?
For ab-initio, you should be relaxing the resulting structures after abinitio in a full-atom scorefunction (score12). Then you want to cluster the structures + rescore them using the full-atom scorefunction, if you don’t have the scores from the relax step. For clustering, I would recommend calibur. Its quick and easy http://sourceforge.net/projects/calibur/
In the C++ application, this is known as Ab-Relax. To do this in PyRosetta, create + run a fastrelax instance after you bring the model to all-atom + repack: