Buccaneer is giving good results on a lot of cases now, and can work at
resolutions down to ~3.7A when the phases are very good. However, when
the starting phases are poor, it struggles.
I've recently been analysing some of the hard cases from my test data -
in particular one case at 3.0 and one at 3.2A.
- In the 3.2A case, on defaults buccaneer only builds and sequences
about 400 of 1200 residues correctly. In this case, it is possible to
get the whole structure by the simple expedient of running many more
cycles - 100 cycles instead of the normal 3 or 5 cycles.
- In the 3.0A case the starting phases were poor - CC of 0.58 after
density modification. In this case, on defaults buccaneer doesn't build
and sequence any chains correctly. Running for 100 cycles leads to about
100 of 1100 residues being sequenced correctly. After 300 cycles this is
up to 900 of 1100. The calculation is still running.
One problem is that progress is very jumpy - sometimes the model gets
worse again before getting better. I don't have a good solution to this
yet. Combined strategies alternating buccaneer with ARP/wARP and/or
phenix.autobuild may be helpful in this area.
For most cases, buccaneer builds everything it is going to build in 2-5
cycles, so running for hundreds of cycles is not a good choice, but for
really borderline problems it can be helpful. (However running extra
cycles can also cause coordinate errors and R-factors to drop, if rather
slowly. A 'quick' 'rebuild' of the buccaneer model in phenix.autobuild
does the same job quicker.)
Kevin
