The SAM T99 and T2K scripts use an A2M alignment as a seed. That alignment is often a single sequence, but it can be any multiple alignment. Upper and lower case letters are used for identifying the alignment columns and insertions respectively ("-" is used for a gap in an alignment column, and "." may be used to pad insertions). If you provide a sequence as a seed that has a large insertion in the middle, the constructed HMM will favor having insertions at that point, and the target99 or target2k script will (usually) construct a pretty good HMM for recognizing the split domain, without any further guidance. Occassionally, the target99 or target2k script will have "model drift" in which the HMM no longer aligns the initial sequence the way we would want---this causes problems in about 1% of the HMMs we build, but may be a bigger problem in split-domain HMMs. The next version of SAM will, I hope, have some new features to fix this problem (I have some ideas on how to fix it, but am waiting for Richard Hughey to implement them, or for a grad student to volunteer to implement them).