Just are a few links I like to share:
... and a few OpenSource/Freeware projects I have been involved with:
BASILISK is a probabilistic model of the conformational space of amino acid side chains in proteins.
Unlike rotamer libraries, BASILISK models the chi angles in continuous space,
including the influence of the protein's backbone.
Phaistos is a collection of tools for proteins structure prediction. It
currently features the FB5DBN and TorusDBN models, which make it
possible to sample protein structures compatible with a given amino
acid and/or secondary structure sequence.
Mocapy++ is a Dynamic Bayesian Network toolkit, implemented in C++. It supports
discrete, multinomial, Gaussian, Kent, Von Mises and Poisson nodes. Inference
and learning is done by Gibbs sampling/Stochastic-EM.