This is all I got back:
“I haven’t made any effort myself to transform the scores from backrub runs into approximate ddGs. What I have done is described in the JMB backrub paper, namely run 10 different 10,000 step simulations at a kT of 0.6 for both the wild type and mutant. I then take the lowest score.
Obviously that’s a pretty simplistic protocol that far undersamples.”