HMMs can be constructed directly from a single sequence, by adding
one state for each character in the sequence and connecting them in a
single path. The model can be made to match sequences of any length
by adding a self-loop edge to the first and last states.
Such a model is not very useful in itself, since it will only match
sequences which are nearly identical, but it can serve as a starting
point for the model modification techniques described in Section 2.8 
.
From a seed with multiple sequences one could also construct parallel paths from the start state to the stop state, one per sequence. Unfortunately, such a model would offer no advantages over far more efficient string search techniques, and would be much too large for seeds of reasonable size.