Posts by bblum

1) Message boards : Number crunching : Problems with Rosetta 5.95 (Message 51992) Posted 17 Mar 2008 by bblum Post: Cristoph is correct: the new diversified beta sampling was added to version 5.94 on Ralph, which never made it to Rosetta@home because of the small bug that we discovered during the course of our tests on Ralph and corrected in version 5.95.
2) Message boards : Number crunching : Problems with Rosetta 5.95 (Message 51937) Posted 14 Mar 2008 by bblum Post: Please post any problems you experience with Rosetta 5.95 here and we will do our best to address them as quickly as possible. Thanks so much for your help.
3) Message boards : Number crunching : Rosetta Application Version Release Log (Message 51936) Posted 14 Mar 2008 by bblum Post: Rosetta version 5.95 This version introduces a new "resampling" method for diversified sampling of beta topologies. It expands the range of conformation space that Rosetta can search by learning promising regions of search space from previous search results.
4) Message boards : Number crunching : Rosetta Application Version Release Log (Message 41742) Posted 2 Jun 2007 by bblum Post: Rosetta@home has been updated to version 5.68. This version allows us to carry out all-atom refinement for really large symmetric proteins where we simulate only subsystems, fixes a small bug in the RNA mode, and will allow a new sort of RNA simulation with lots of imposed base pairs.
5) Message boards : Number crunching : Problems with Rosetta version 5.68 (Message 41741) Posted 2 Jun 2007 by bblum Post: Hi all. Please let us know here if any issues arise with the new Rosetta version.
6) Message boards : Rosetta@home Science : Rosetta@home Active WorkUnit(s) Log (Message 22497) Posted 15 Aug 2006 by bblum Post: I've added three new jobs to the queue: 1di2__LARS_ABRELAX_SAVE_ALL_OUT_BARCODE_ 1di2__CHEAT_ABRELAX_SAVE_ALL_OUT_BARCODE_ 1di2__CONTROL_ABRELAX_SAVE_ALL_OUT_ All are "abrelax" runs (ab initio prediction followed by relaxation, as described below by Bin and Rhiju--graphically speaking, big wiggles followed by small wiggles). The protein is 1di2A. We've already generated a population of decoys for 1di2A, and now we're attempting to use what we've learned from that population to generate a much better population. It's frequently clear from an initial sampling run that certain decoy features are always good--they tend, if present, to give rise to low energy. If we can identify these features, then we can concentrate the next round of sampling on them. We've generally found that features which correlate with low energy in the initial sampling round tend to be present in the native. Features can include both low-level ones (e.g. torsion angles or rotamers for specific residues) and high-level ones (e.g. beta strand pairings). In these jobs, we're only enriching for low-level features. The job I'm most excited about is the LARS job; here we've used a sparse linear model to identify a few features that correlated very strongly with the Rosetta energy, and I've enriched for all of them in sampling. The CHEAT job fixes a single linchpin native feature that we hope will result in sampling much closer to the native; it will allow us to test a hypothesis we have about the joint distribution of features in Rosetta sampling. The CONTROL job fixes no features and will serve as a baseline.