Mark: performance enhancement: found solution to reduce requirement from quadratic to linear: a not too well know paper from Guan and Uberbacher describes: instead of picking one point in the middle, do this in a forward run, keep one tracepoint and keep updating it as you continue on through scoring and you have the point chosen that is on the path of the best score for this row.
next thought : you can do that for multiple rows, do 10 points each pass and you end up with linear space requirement. Gets rid of overhead of Myers&Miller with only 10% increase in time component.
Scooter sent out an email with a comparison of a few alignments. The alignments look very strange to us. Did the alignment get created with gap symbols as part of the sequence? Both BioJava3 and jaligner seem to show odd behaviour. Mark will investigate this and also how different gap penalties influence this alignment.
End is approaching, we will have a last conference call on Tuesday and discuss final request for now.