My Bayes Net Toolbox gives you access to all its internal data structures, including join trees, so you can look at the clique sizes etc. The graph theory routines are completely separate from the rest of the toolbox: they just take as input an adjacency matrix, which can be loaded in from a file. http://HTTP.CS.Berkeley.EDU/~murphyk/Bayes/bnt.html Kevin P.S. I have recently added MCMC structure learning to BNT (currently restricted to fully observed, discrete, static networks), which several people had requested (I forget exactly who...).
