Wed Aug 14 14:46:42 PDT 2002
t0185

Homology modeling 1uag, 1e8cA, ... 
	c.59.1.1, c.72.2.1

1 Sept 2002 Kevin Karplus 08:18

Top model currently try1-opt-scwrl.
Raising burial and break weights leaves try1-opt-scwrl at the top.

Some of the good parts of the T0185-1uag alignment seem to have been
lost---we may need to extract some hbond constraints from that alignment.

Check 2ry prediction for D377-I386, which is predicted to be helix but
aligned to a strand.

2 Sept 2002 14:19 Kevin Karplus

D377-386 is a very weak 2ry prediction.

Perhaps we need to break at the domain boundaries in the alignment
with 1uag (say at E82 or K98 and at V297) and make separate alignments for
each subdomain.

This won't help any with the weak predictions in the region S351-K400,
so maybe we should pull out that region to get its own alignments and
fragments also.

OK--let's do
	t0185-1-100
	t0185-81-300
	t0185-301-457
	t0185-351-400
as "subdomains" to get their own alignments.

16:26 
	The subdomain makes are done, and all but the tiny 351-400
	still like 1uag best, but there are more alignments to sample now.


Rerunning from scratch.

Thu Sep  5 13:58:48 PDT 2002

try2-opt is not bad---we need to add a few fairly obvious constraints
to reattach pieces which have drifted off a little:
	The first domain of 1uag has	5^4^1^2^3^
	and we have all these strands.
	
	The second domain has 	14^13v12^11^10^9^6^8^7^
	
			
	L206-211	parallel to N171-176 and T230-R236
	
	V133-G136 or M139-E143 next to P153-E157
		(look at 1uag to determine strand order)


	The third domain has 
		15v16^20^17^18^19^

	strand K340-I344 parallel to Y313-D317
	(or should it be separated as in the 3rd domain of 1uag, wiht
	Y313-D317 corresponding to 16 and K340-I34 corresponding to 17.
	
	E369-E381 corresponds to 18
	L398-Y403 corresponds to 19
	T421-R299 corresponds to 20
			

Foo---too much thinking!  I'll just pick up Hbonds from
T0185-1uag.pdb, clean them up a bit, and use them for
try3.constraints. 

Thu Sep  5 16:16:39 PDT 2002

I think I screwed up a bit and started the try3 run with an old
version of undertaker.script, so some of its results are incorrectly
labeled as being from try2.  I've kiled it and will start a try4 run.

Fri Sep  6 11:20:23 PDT 2002

Current best is try4-opt.  Little problems:	
	L255-K258 wound into a helix, but predicted to be
		strand---this looks like it could be straightened,
		perhaps shifting register of K244-K249 to line up K249
		and R253.
		
	230>     tYTLEMRSAS rae
	250<       nKEVMAKq
	251>      gKRYLELKLkvp
		
	
	L142-Y147 This looks like it could be fixed by attaching 
		E143-Y147 anti-parallel to P131-L135
	
	130>     sPTVFlgg
	150<   kgeYNGHElds

17:39

	try5-opt has cleaned up the sheet around 250, but still has
	142-147 in a helix.

21:48

	try6-opt STILL has 142-147 in a helix.
	I'll try adding some Hbond constraints for it and up
	constraint weight in the scoring function.
	I'll also turn OptSubtree down somewhat, and turn up the
	reshaping operators, since the problem is not one of getting
	the pieces in the right places, but of reshaping the pieces.