CHANGE the blastall in the Makefile for recent.blast to use 
blastall -F -m 9


Fix the conversion from eps to pdf so that long sequence logos are not truncated.


Fix the sequences from kumar.seqs (the target sequences have a * at
the end, though this has been fixed in the kumar.seqs file).
Alternatively, fix undertaker to accept and ignore a * character in a sequence.

Change Make.main and associated scripts to use latest neural nets and
scripts for building web pages (see pce/protein-predict/SAM_T05test)

Revise Makefile to use a more selective search for targets to
redo---BLAST e-value should be no worse than 100 time the E-value
recorded in new-diagonal.rdb.



DONE:

<P>We need to devise an update procedure for updating the predictions
periodically as new sequences and structures are added to the
underlying databases.
For example, we could score the yeast sequences with each new addition
to the template library.
If any of the yeast sequences score well, we could mark them for
re-prediction.
Once a month we could re-do the predictions for the marked sequences
based on the new template library, perhaps using the existing t2k
alignments, perhaps generating new t2k alignments.
There are other ways we could trigger re-prediction---manual request
by users (to be confirmed by us, to avoid massive recomputations),
BLAST searches with new PDB sequences as keys, ... .
The find-recent-ids script in /projects/compbio/data/pdbfinder may be
useful in selecting the subset of pdb to be considered new.
Big chunks of the procedure have been put into the Makefile in the
main yeast directory (see the URL above or look in
/projects/compbio/experiments/protein-predict/yeast/Makefile),
including extracting the recent PDB sequences, blasting against the
yeast ORFs, and collecting the hits as an IDs file.  I still have not
automated the decision about what constitutes "recent"
nor the
submitting of ids to be searched again.  This is all pretty
easy---a bigger problem is figuring out what to do about the delays
between PDB releasing a structure and it appearing in the PDBAA
non-redundant list that we search (about a month) or our template
library (a week to a month).  Based on a search done at the end of
December 2002, it looks like we'll need to redo about 60 searches a
month---about 1% of the database.

<P>I've more or less automated the update of the yeast predictions
now, so there isn't much to do here. [April 2003]


<P>
Once a year we should probably update the t2k alignments.
Rather than having a massive annual update, it might be better to have
a rolling update, redoing the alignments a few at a time (about 2% a
week), oldest alignments first.
Note that this maintenance is much more expensive than the "recent
PDB" maintenance, both per sequence and in number of sequences.

<P>
When we do a reprediction, we would have to make sure that whatever
database is used for indexing the results is properly updated.
Having some timestamps for the t2k alignment and the structure
prediction in the database would probably help in devising a good
update procedure.