On the use of pseudo-likelihoods in Bayesian variable selection