In the fold recognition approach to structure prediction,
a sequence is tested for compatibility with an already
known fold. For membrane proteins, however, few folds have
been determined experimentally. Here the feasibility of
computing the vast majority of likely membrane protein
folds is tested. The results indicate that conformation
space can be effectively sampled for small numbers of helices.
The vast majority of potential monomeric membrane protein
structures can be represented by about 30-folds for three
helices, but increases exponentially to about 1,500,000
folds for seven helices. The generated folds could serve
as templates for fold recognition or as starting points
for conformational searches that are well distributed throughout
conformation space.