> As a general argument, just because an alignment has more true matches > does not mean that it is correct. There are penalties associated with > those two gaps you introduced that offset the gain from exact residue > matching. Not knowing exactly which sequence you are modeling, I > submitted P11229 (ACM1_HUMAN) to the SUPERFAMILY server and the > alignment come back with no gaps in that region. Take a look at: > > http://www.supfam.org/SUPERFAMILY/cgi-bin/gene.cgi?genome=up;seqid=P11229 > > and follow the "Align" link. Obviously, the same can be done for any > protein for which you know the SwissProt number by simply inserting it > at the end of the link above. Not to get into prolonged argument here, > I'll just say that I would trust SUPERFAMILY-based alignment over any > human-derived alignment unless there is a very good reason to believe > that the server alignment is wrong. And breaking a long helix to make > few extra positive matches usually doesn't qualify as a good reason, > especially because it forces you to make a compensatory two-residue > gap later in the model. But don't take my word for it, try modeling > the protein with and without those gaps and see which model evaluates > better. > > Hope this helps, > > Mensur Here there are strong argument that kinda end the gap debate ... Thanks for preventing me to lost my time and make huge mistakes. (the gapped models came with objective function score at least double of the 'classical' ones)
At least have I learned to play with the alpha constraint of modeller ;o) .
Regards
Guillaume