On Thu, Nov 18, 2010 at 5:47 PM, Nathaniel Echols wrote:

On Thu, Nov 18, 2010 at 2:31 PM, Bryan Lepore wrote: I think it is expecting exactly the same output as CLUSTALW, which has a slightly different first >line than MAFFT.

bingo

 We now distribute a multiple sequence alignment program >called MUSCLE with Phenix - you can use it like this:

yes thanks - i saw that in there ;^)

Sculptor should also take alignments in FASTA or PIR format.  (Make sure you use the correct file extension,

HERE we go - .fasta is running. but clustalw extension is .aln, right? oh also : wouldn't this be a good candidate for multi-processor threading? -Bryan

oh, and i figure that the alignment needs the structure's sequence in there - right? this would actually save me some keystrokes. -Bryan

Yes, and the target sequence (i.e. the protein that you crystallized) needs to be first. One annoying thing about MUSCLE is that it sorts sequences and (currently) won't let you disable this behavior, so you may need to do some cut-and-pasting. Maybe we need to modify the Python wrapper to do this automatically. (Although I think it would be even easier if Sculptor did the alignment automatically.)

-Nat

This is actually a bit odd, since one could run version 3.7 of MUSCLE with a -stable flag that would output sequences in the the same order as the input... but apparently this has been disabled in the version that ships with PHENIX (3.8.31). Maybe we could just ask Robert Edgar to put it back? Luca

But I still think the solution is to do the alignment internally - the user should be able to provide a PDB file and a target sequence, and the program will run MUSCLE automatically and figure out which row in the alignment output corresponds to each sequence. This avoids an extra step and would clear up the confusion regarding formats, ordering, etc.

-Nat

I agree this would be a solution to at least part of the problem. There is already some support for this, e.g. sequence { file_name = foo.seq chain_ids = A,B,C } would create an alignment internally using the sequence from foo.seq and chains A, B and C (and apply it to the same chains). It is not currently using muscle, but there is possibly no need - these are only pairwise sequence alignments. On the other hand, the alignment can be of crucial importance, and the user should not be limited to alignment algorithms implemented in Sculptor (and it is possibly not realistic to have an interface to all alignment programs, many of which are web services). Creating multiple sequence alignments via Sculptor (which may be more precise) would also complicate the user interface, for little apparent gain. BW, Gabor

On Fri, Nov 19, 2010 at 5:16 AM, Dr G. Bunkoczi wrote:

Would it be sufficient to provide an option, so that one can specify the index of the target sequence in the alignment file?

in the interest of automation, i guess - though the titles (the stuff on the line starting with ">") can get pretty disgusting if you keep everything from the database it came from : non-alphabetic characters, spaces,

Would it be better to provide the target sequence, and let Sculptor figure out the index itself?

can't say if its better, but if it works - why not. i have seen programs where you say e.g. "fubar" and then the alignment has a part with ">fubar" then everything between ">fubar" and the next ">" is defined as the target sequence (IIUC). ... anyways, how is the target sequence used that a consensus sequence couldn't be used instead? i mean, the model looks like it has one applied to it. -Bryan