Rosetta all-atom decoys, a test set.

Updated by Rhiju Das: Oct. 15, 2007
Posted on the web by Rhiju Das: Oct. 15, 2007


This directory contains well-scoring Rosetta protein models and their native crystal structures for 
59 proteins. [There were 62 at some point, but 3 were removed due to the native structure
being solved by NMR]. The subdirectories are:

natives           : crystal structures, in some cases truncated from PDB entries.
relaxed_natives   : crystal structures that have been put through refinement in Rosetta (relieving
		        clashes and, in some cases, repacking rotamers)
low_score_decoys  : 100 lowest scoring models out of ~10,000 total models, using the Rosetta de novo 
		        structure prediction algorithm followed by all-atom refinement.  
first_decoys      : 20 random models, no score cut, useful for compiling rotamer statistics, etc.

This is a useful snapshot of the "state-of-the-art" in Rosetta structure prediction.

Caveats!

1. One great use for this set is to calibrate and validate new algorithms that may discriminate 
   the native state from the Rosetta "decoys". For that purpose, it is critical to take into account that
   some of the decoys are *already* close to the native state.

2. The low scoring decoys have been produced with a "full-atom feedback" algorithm (R. Das, 
   D. Baker, unpublished results) that, in several cases, guides low-resolution fragment assembly to 
   conformations that score better after all-atom refinement. 

3. Some of the directories have linked ".out" files; please ignore them.

Please feel free to e-mail rhiju @ u.washington.edu with any questions.






