There is still lots to do for casp6:

1) make a script that converts best-models.pdb.gz into a legal CASP6 submission.
	[DONE 20 June 2004]

2) make a script that generates HelixConstraint and StrandConstraint
	commands from the secondary structure prediction.  (Only the
	strongest predictions should be used.)
	[DONE 8 June 2004]

3) document the procedures for starting a new target.
	[partly done 7 June 2004---see README]

4) add Rosetta repacking as an optional step.
	Needs scripts to generate the .res file to have NATRO for
	proline and cysteine.  (At least for disulfide cys.)
	May want to do this from within undertaker as a
	RosettaPackConform command, similar to the SCWRLConform command.
	
	[DONE with scripts 17 June 2004
	may not be worth putting into undertaker yet
	]

5) add the new neural nets to the Make.main file for prediction using
   the T04 alignments.
   	[Done 15 June 2004]
   
6) add profile-profile alignments to the pool of alignments for
   templates in the library (both muscle and SAM-based alignments).
   This is mostly a Makefile task.  
   
   [MUSCLE profile-profile done, 15 June 2004]
   Still to do: SAM profile-profile.

7) Maybe switch from using dunbrack-1332 as training set to dunbrack-50pc-2621
   for the undertaker runs.
   (First check for any bad structures in the new set, removing any
   that are very short or consistently mispredicted by the neural nets).

8) Do a better job picking the best initial alignments, so that we have
   better starting points for undertaker.  Are we better off with many
   slightly different alignments or with one alignment that is more
   likely to be right?  (That is, how good is undertaker at picking
   out the right alignment from a pool of alignments.)

9) start building a web server (like the SAM-T02 one) for SAM-T04

10) Make all servers run under parasol instead of condor
	[DONE 13 June 2004 Kevin Karplus]

11) Modify superimpose-best.under based on conserved residues, so that
	superposition starts with the most conserved residues.