Learning a causal structure: a Bayesian Random Graph approach