more on casp

Part of the reason for my recent hiatus at writing entries is a combination of attending meetings and finishing up with semester business.  Anyway, I have some interesting results from casp-8.

  1. We did pretty well on the targets where there was one good template and we found it.  (by some measures we were #1 and #2 on a couple of targets).  This is good because high-quality molecular mechanics is a major research focus for my group.  Our potential probably slightly enlarges protein models and one of the “tricks of casp” is to compress your model.
  2. Our server was not designed to handle many simultaneous hits and one of the neat features of profile-profile methods is the number of homologs that they can fish out.  Usually if it finds one, it finds 20.  The problem is which combination of these 20 is the best model.
  3. It is critical to keep the profiles up to date.  Psi-blast has its problems and one of them is non-stationarity.  If you run on different databases then you get radically different profiles – and with updates over time the profiles sort of “rot”.   (Its other problems include some rather poor software engineering and a context dependence).  It doesn’t run in my hands with the non-redundant database (nr) but does work much better with the refseq_protein database.  This caused a number of problems where we missed the best profiles.  ARRGH!
  4. It is probably a good idea to move from a single geometric measure (gdt and related variations) to explaining the data.  For NMR this means can your model reproduce the NOE’s and backbone restraints used to produce the “structure”.  For crystallography, it is probably good to ask what is the height of the cross-rotation function.  I’m debating building a server for this.  There are more than one group that does very well by having trained their method to generate models with poor geometry (useless models) that generate good scores on the geometric measures.  Forcing evaluation against data would fix this issue as well as be statistically and mathematically the best measure.  – you can argue about measures but you can’t argue against data.
Written by Rob in: Uncategorized |

No Comments »

RSS feed for comments on this post. TrackBack URL

Leave a comment

Powered by WordPress | Aeros Theme | WordPress Themes